参数设置

/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug/bin/main
options:-h,        --help              [default] show this help message and exit-t N,      --threads N         [4      ] number of threads to use during computation-p N,      --processors N      [1      ] number of processors to use during computation-ot N,     --offset-t N        [0      ] time offset in milliseconds-on N,     --offset-n N        [0      ] segment index offset-d  N,     --duration N        [0      ] duration of audio to process in milliseconds-mc N,     --max-context N     [-1     ] maximum number of text context tokens to store-ml N,     --max-len N         [0      ] maximum segment length in characters-sow,      --split-on-word     [false  ] split on word rather than on token-bo N,     --best-of N         [5      ] number of best candidates to keep-bs N,     --beam-size N       [5      ] beam size for beam search-wt N,     --word-thold N      [0.01   ] word timestamp probability threshold-et N,     --entropy-thold N   [2.40   ] entropy threshold for decoder fail-lpt N,    --logprob-thold N   [-1.00  ] log probability threshold for decoder fail-debug,    --debug-mode        [false  ] enable debug mode (eg. dump log_mel)-tr,       --translate         [false  ] translate from source language to english-di,       --diarize           [false  ] stereo audio diarization-tdrz,     --tinydiarize       [false  ] enable tinydiarize (requires a tdrz model)-nf,       --no-fallback       [false  ] do not use temperature fallback while decoding-otxt,     --output-txt        [false  ] output result in a text file-ovtt,     --output-vtt        [false  ] output result in a vtt file-osrt,     --output-srt        [false  ] output result in a srt file-olrc,     --output-lrc        [false  ] output result in a lrc file-owts,     --output-words      [false  ] output script for generating karaoke video-fp,       --font-path         [/System/Library/Fonts/Supplemental/Courier New Bold.ttf] path to a monospace font for karaoke video-ocsv,     --output-csv        [false  ] output result in a CSV file-oj,       --output-json       [false  ] output result in a JSON file-ojf,      --output-json-full  [false  ] include more information in the JSON file-of FNAME, --output-file FNAME [       ] output file path (without file extension)-ps,       --print-special     [false  ] print special tokens-pc,       --print-colors      [false  ] print colors-pp,       --print-progress    [false  ] print progress-nt,       --no-timestamps     [false  ] do not print timestamps-l LANG,   --language LANG     [en     ] spoken language ('auto' for auto-detect)-dl,       --detect-language   [false  ] exit after automatically detecting language--prompt PROMPT     [       ] initial prompt-m FNAME,  --model FNAME       [models/ggml-base.en.bin] model path-f FNAME,  --file FNAME        [       ] input WAV file path-oved D,   --ov-e-device DNAME [CPU    ] the OpenVINO device used for encode inference-ls,       --log-score         [false  ] log best decoder scores of tokens-ng,       --no-gpu            [false  ] disable GPU

调试设置

项目依赖和CmakeLists.txt

set(TARGET main)
add_executable(${TARGET} main.cpp)include(DefaultTargetOptions)target_link_libraries(${TARGET} PRIVATE common whisper ${CMAKE_THREAD_LIBS_INIT})

#include "common.h"#include "whisper.h"#include <cmath>
#include <fstream>
#include <cstdio>
#include <string>
#include <thread>
#include <vector>
#include <cstring>

main

int main(int argc, char ** argv) {// 1.解析参数whisper_params params;// 解析命令行参数，将结果保存到params中if (whisper_params_parse(argc, argv, params) == false) {… }// 检查输入文件名是否为空if (params.fname_inp.empty()) {… }// std::vector<std::string> fname_inp = {};// 检查语言参数是否有效if (params.language != "auto" && whisper_lang_id(params.language.c_str()) == -1) {… }// 检查两个布尔参数，如果同时为真，执行相应的错误处理代码if (params.diarize && params.tinydiarize) {… }// whisper initstruct whisper_context_params cparams;cparams.use_gpu = params.use_gpu;// 2.使用whisper初始化上下文，并根据给定的模型文件和参数进行配置struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);if (ctx == nullptr) {fprintf(stderr, "error: failed to initialize whisper context\n");return 3;}// initialize openvino encoder. this has no effect on whisper.cpp builds that don't have OpenVINO configured// 初始化OpenVINO编码器，对于没有配置OpenVINO的whisper.cpp构建，此调用无效whisper_ctx_init_openvino_encoder(ctx, nullptr, params.openvino_encode_device.c_str(), nullptr);// 3.对输入文件列表进行循环处理for (int f = 0; f < (int) params.fname_inp.size(); ++f) {… }whisper_print_timings(ctx); // 打印whisper上下文的计时信息whisper_free(ctx);// 释放whisper上下文占用的资源return 0;
}

1.解析参数

• 略

2.使用whisper初始化上下文，并根据给定的模型文件和参数进行配置

• webassembly003 whisper.cpp的main项目-2:根据给定的模型文件和参数进行配置

3.对输入文件列表进行循环处理

3.1解析参数

        const auto fname_inp = params.fname_inp[f]; // "/home/***/whisper.cpp-1.5.0/samples/jfk.wav"const auto fname_out = f < (int) params.fname_out.size() && !params.fname_out[f].empty() ? params.fname_out[f] : params.fname_inp[f]; // "/home/***/whisper.cpp-1.5.0/samples/jfk.wav"

3.2根据参数读取音频

        std::vector<float> pcmf32;               // mono-channel  单声道(音频只有一个声道) ,采样点类型为32位浮点数, `PCM` 表示脉冲编码调制std::vector<std::vector<float>> pcmf32s; // stereo-channel 立体声，即音频有两个声道（左声道和右声道）// read_wav 定义在 common.cpp, 如果在函数调用之前使用::，并且没有指定任何命名空间，那么它会被解释为全局命名空间。if (!::read_wav(fname_inp, pcmf32, pcmf32s, params.diarize)) { // if (!::read_wav(...))：使用 if 语句检查读取 WAV 文件的结果。! 表示逻辑取反，所以如果 read_wav 返回 false（表示读取失败），则执行下面的代码块。fprintf(stderr, "error: failed to read WAV file '%s'\n", fname_inp.c_str());continue;}

3.3print information

        // print system information{fprintf(stderr, "\n");fprintf(stderr, "system_info: n_threads = %d / %d | %s\n",params.n_threads*params.n_processors, std::thread::hardware_concurrency(), whisper_print_system_info());}// print some info about the processing{fprintf(stderr, "\n");if (!whisper_is_multilingual(ctx)) {if (params.language != "en" || params.translate) {params.language = "en";params.translate = false;fprintf(stderr, "%s: WARNING: model is not multilingual, ignoring language and translation options\n", __func__);}}if (params.detect_language) {params.language = "auto";}fprintf(stderr, "%s: processing '%s' (%d samples, %.1f sec), %d threads, %d processors, %d beams + best of %d, lang = %s, task = %s, %stimestamps = %d ...\n",__func__, fname_inp.c_str(), int(pcmf32.size()), float(pcmf32.size())/WHISPER_SAMPLE_RATE,params.n_threads, params.n_processors, params.beam_size, params.best_of,params.language.c_str(),params.translate ? "translate" : "transcribe",params.tinydiarize ? "tdrz = 1, " : "",params.no_timestamps ? 0 : 1);fprintf(stderr, "\n");}

3.4run the inference

3.4.1解析参数

            whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);wparams.strategy = params.beam_size > 1 ? WHISPER_SAMPLING_BEAM_SEARCH : WHISPER_SAMPLING_GREEDY;wparams.print_realtime   = false;wparams.print_progress   = params.print_progress;wparams.print_timestamps = !params.no_timestamps;wparams.print_special    = params.print_special;wparams.translate        = params.translate;wparams.language         = params.language.c_str();wparams.detect_language  = params.detect_language;wparams.n_threads        = params.n_threads;wparams.n_max_text_ctx   = params.max_context >= 0 ? params.max_context : wparams.n_max_text_ctx;wparams.offset_ms        = params.offset_t_ms;wparams.duration_ms      = params.duration_ms;wparams.token_timestamps = params.output_wts || params.output_jsn_full || params.max_len > 0;wparams.thold_pt         = params.word_thold;wparams.max_len          = params.output_wts && params.max_len == 0 ? 60 : params.max_len;wparams.split_on_word    = params.split_on_word;wparams.speed_up         = params.speed_up;wparams.debug_mode       = params.debug_mode;wparams.tdrz_enable      = params.tinydiarize; // [TDRZ]wparams.initial_prompt   = params.prompt.c_str();wparams.greedy.best_of        = params.best_of;wparams.beam_search.beam_size = params.beam_size;wparams.temperature_inc  = params.no_fallback ? 0.0f : wparams.temperature_inc;wparams.entropy_thold    = params.entropy_thold;wparams.logprob_thold    = params.logprob_thold;whisper_print_user_data user_data = { &params, &pcmf32s, 0 };// this callback is called on each new segmentif (!wparams.print_realtime) {wparams.new_segment_callback           = whisper_print_segment_callback;wparams.new_segment_callback_user_data = &user_data;}if (wparams.print_progress) {wparams.progress_callback           = whisper_print_progress_callback;wparams.progress_callback_user_data = &user_data;}

whisper_print_segment_callback:获取片段的推理结果并打印的回调函数

void whisper_print_segment_callback(struct whisper_context * ctx, struct whisper_state * /*state*/, int n_new, void * user_data) {const auto & params  = *((whisper_print_user_data *) user_data)->params;const auto & pcmf32s = *((whisper_print_user_data *) user_data)->pcmf32s;const int n_segments = whisper_full_n_segments(ctx);std::string speaker = "";int64_t t0 = 0;int64_t t1 = 0;// print the last n_new segmentsconst int s0 = n_segments - n_new;if (s0 == 0) {printf("\n");}for (int i = s0; i < n_segments; i++) {if (!params.no_timestamps || params.diarize) {t0 = whisper_full_get_segment_t0(ctx, i);t1 = whisper_full_get_segment_t1(ctx, i);}if (!params.no_timestamps) {printf("[%s --> %s]  ", to_timestamp(t0).c_str(), to_timestamp(t1).c_str());}if (params.diarize && pcmf32s.size() == 2) {speaker = estimate_diarization_speaker(pcmf32s, t0, t1);}if (params.print_colors) {for (int j = 0; j < whisper_full_n_tokens(ctx, i); ++j) {if (params.print_special == false) {const whisper_token id = whisper_full_get_token_id(ctx, i, j);if (id >= whisper_token_eot(ctx)) {continue;}}const char * text = whisper_full_get_token_text(ctx, i, j);const float  p    = whisper_full_get_token_p   (ctx, i, j);const int col = std::max(0, std::min((int) k_colors.size() - 1, (int) (std::pow(p, 3)*float(k_colors.size()))));printf("%s%s%s%s", speaker.c_str(), k_colors[col].c_str(), text, "\033[0m");}} else {const char * text = whisper_full_get_segment_text(ctx, i);printf("%s%s", speaker.c_str(), text);}if (params.tinydiarize) {if (whisper_full_get_segment_speaker_turn_next(ctx, i)) {printf("%s", params.tdrz_speaker_turn.c_str());}}// with timestamps or speakers: each segment on new lineif (!params.no_timestamps || params.diarize) {printf("\n");}fflush(stdout);}
}

3.4.2解析参数

            // examples for abort mechanism// in examples below, we do not abort the processing, but we could if the flag is set to true// the callback is called before every encoder run - if it returns false, the processing is aborted{static bool is_aborted = false; // NOTE: this should be atomic to avoid data racewparams.encoder_begin_callback = [](struct whisper_context * /*ctx*/, struct whisper_state * /*state*/, void * user_data) {bool is_aborted = *(bool*)user_data;return !is_aborted;};wparams.encoder_begin_callback_user_data = &is_aborted;}// the callback is called before every computation - if it returns true, the computation is aborted{static bool is_aborted = false; // NOTE: this should be atomic to avoid data racewparams.abort_callback = [](void * user_data) {bool is_aborted = *(bool*)user_data;return is_aborted;};wparams.abort_callback_user_data = &is_aborted;}

3.4.3 process whisper_full_parallel

            if (whisper_full_parallel(ctx, wparams, pcmf32.data(), pcmf32.size(), params.n_processors) != 0) {fprintf(stderr, "%s: failed to process audio\n", argv[0]);return 10;}

whisper_full_parallel

int whisper_full_parallel(struct whisper_context * ctx,struct whisper_full_params params,const float * samples,int n_samples,int n_processors) {if (n_processors == 1) {return whisper_full(ctx, params, samples, n_samples);}// 略
}

whisper_full

int whisper_full(struct whisper_context * ctx,struct whisper_full_params   params,const float * samples,int   n_samples) {return whisper_full_with_state(ctx, ctx->state, params, samples, n_samples);
}

whisper_full_with_state(推理的关键代码800行)

//没细看，有空再看
int whisper_full_with_state(struct whisper_context * ctx,struct whisper_state * state,struct whisper_full_params   params,const float * samples,int   n_samples) {// clear old resultsauto & result_all = state->result_all;result_all.clear();if (n_samples > 0) {// compute log mel spectrogramif (params.speed_up) {// TODO: Replace PV with more advanced algorithmWHISPER_LOG_ERROR("%s: failed to compute log mel spectrogram\n", __func__);return -1;} else {if (whisper_pcm_to_mel_with_state(ctx, state, samples, n_samples, params.n_threads) != 0) {WHISPER_LOG_ERROR("%s: failed to compute log mel spectrogram\n", __func__);return -2;}}}// auto-detect language if not specifiedif (params.language == nullptr || strlen(params.language) == 0 || strcmp(params.language, "auto") == 0 || params.detect_language) {std::vector<float> probs(whisper_lang_max_id() + 1, 0.0f);const auto lang_id = whisper_lang_auto_detect_with_state(ctx, state, 0, params.n_threads, probs.data());if (lang_id < 0) {WHISPER_LOG_ERROR("%s: failed to auto-detect language\n", __func__);return -3;}state->lang_id = lang_id;params.language = whisper_lang_str(lang_id);WHISPER_LOG_INFO("%s: auto-detected language: %s (p = %f)\n", __func__, params.language, probs[whisper_lang_id(params.language)]);if (params.detect_language) {return 0;}}if (params.token_timestamps) {state->t_beg    = 0;state->t_last   = 0;state->tid_last = 0;if (n_samples > 0) {state->energy = get_signal_energy(samples, n_samples, 32);}}const int seek_start = params.offset_ms/10;const int seek_end = params.duration_ms == 0 ? whisper_n_len_from_state(state) : seek_start + params.duration_ms/10;// if length of spectrogram is less than 1.0s (100 frames), then return// basically don't process anything that is less than 1.0s// see issue #39: https://github.com/ggerganov/whisper.cpp/issues/39if (seek_end < seek_start + (params.speed_up ? 50 : 100)) {return 0;}// a set of temperatures to use// [ t0, t0 + delta, t0 + 2*delta, ..., < 1.0f + 1e-6f ]std::vector<float> temperatures;if (params.temperature_inc > 0.0f) {for (float t = params.temperature; t < 1.0f + 1e-6f; t += params.temperature_inc) {temperatures.push_back(t);}} else {temperatures.push_back(params.temperature);}// initialize the decodersint n_decoders = 1;switch (params.strategy) {case WHISPER_SAMPLING_GREEDY:{n_decoders = params.greedy.best_of;} break;case WHISPER_SAMPLING_BEAM_SEARCH:{n_decoders = std::max(params.greedy.best_of, params.beam_search.beam_size);} break;};n_decoders = std::max(1, n_decoders);if (n_decoders > WHISPER_MAX_DECODERS) {WHISPER_LOG_ERROR("%s: too many decoders requested (%d), max = %d\n", __func__, n_decoders, WHISPER_MAX_DECODERS);return -4;}// TAGS: WHISPER_DECODER_INITfor (int j = 1; j < n_decoders; j++) {auto & decoder = state->decoders[j];decoder.sequence.tokens.reserve(state->decoders[0].sequence.tokens.capacity());decoder.probs.resize   (ctx->vocab.n_vocab);decoder.logits.resize  (ctx->vocab.n_vocab);decoder.logprobs.resize(ctx->vocab.n_vocab);decoder.logits_id.reserve(ctx->model.hparams.n_vocab);decoder.rng = std::mt19937(0);}// the accumulated text context so farauto & prompt_past = state->prompt_past;if (params.no_context) {prompt_past.clear();}// prepare prompt{std::vector<whisper_token> prompt_tokens;// initial promptif (!params.prompt_tokens && params.initial_prompt) {prompt_tokens.resize(1024);prompt_tokens.resize(whisper_tokenize(ctx, params.initial_prompt, prompt_tokens.data(), prompt_tokens.size()));params.prompt_tokens   = prompt_tokens.data();params.prompt_n_tokens = prompt_tokens.size();}// prepend the prompt tokens to the prompt_pastif (params.prompt_tokens && params.prompt_n_tokens > 0) {// parse tokens from the pointerfor (int i = 0; i < params.prompt_n_tokens; i++) {prompt_past.push_back(params.prompt_tokens[i]);}std::rotate(prompt_past.begin(), prompt_past.end() - params.prompt_n_tokens, prompt_past.end());}}// overwrite audio_ctx, max allowed is hparams.n_audio_ctxif (params.audio_ctx > whisper_n_audio_ctx(ctx)) {WHISPER_LOG_ERROR("%s: audio_ctx is larger than the maximum allowed (%d > %d)\n", __func__, params.audio_ctx, whisper_n_audio_ctx(ctx));return -5;}state->exp_n_audio_ctx = params.audio_ctx;// these tokens determine the task that will be performedstd::vector<whisper_token> prompt_init = { whisper_token_sot(ctx), };if (whisper_is_multilingual(ctx)) {const int lang_id = whisper_lang_id(params.language);state->lang_id = lang_id;prompt_init.push_back(whisper_token_lang(ctx, lang_id));if (params.translate) {prompt_init.push_back(whisper_token_translate(ctx));} else {prompt_init.push_back(whisper_token_transcribe(ctx));}}// distilled models require the "no_timestamps" token{const bool is_distil = ctx->model.hparams.n_text_layer == 2;if (is_distil && !params.no_timestamps) {WHISPER_LOG_WARN("%s: using distilled model - forcing no_timestamps\n", __func__);params.no_timestamps = true;}}if (params.no_timestamps) {prompt_init.push_back(whisper_token_not(ctx));}int seek = seek_start;std::vector<whisper_token> prompt;prompt.reserve(whisper_n_text_ctx(ctx));struct beam_candidate {int decoder_idx;int seek_delta;bool has_ts;whisper_sequence sequence;whisper_grammar grammar;};std::vector<std::vector<beam_candidate>> bc_per_dec(n_decoders);std::vector<beam_candidate> beam_candidates;// main loopwhile (true) {if (params.progress_callback) {const int progress_cur = (100*(seek - seek_start))/(seek_end - seek_start);params.progress_callback(ctx, ctx->state, progress_cur, params.progress_callback_user_data);}// of only 1 second left, then stopif (seek + 100 >= seek_end) {break;}if (params.encoder_begin_callback) {if (params.encoder_begin_callback(ctx, state, params.encoder_begin_callback_user_data) == false) {WHISPER_LOG_ERROR("%s: encoder_begin_callback returned false - aborting\n", __func__);break;}}// encode audio features starting at offset seekif (!whisper_encode_internal(*ctx, *state, seek, params.n_threads, params.abort_callback, params.abort_callback_user_data)) {WHISPER_LOG_ERROR("%s: failed to encode\n", __func__);return -6;}// if there is a very short audio segment left to process, we remove any past prompt since it tends// to confuse the decoder and often make it repeat or hallucinate stuffif (seek > seek_start && seek + 500 >= seek_end) {prompt_past.clear();}int best_decoder_id = 0;for (int it = 0; it < (int) temperatures.size(); ++it) {const float t_cur = temperatures[it];int n_decoders_cur = 1;switch (params.strategy) {case whisper_sampling_strategy::WHISPER_SAMPLING_GREEDY:{if (t_cur > 0.0f) {n_decoders_cur = params.greedy.best_of;}} break;case whisper_sampling_strategy::WHISPER_SAMPLING_BEAM_SEARCH:{if (t_cur > 0.0f) {n_decoders_cur = params.greedy.best_of;} else {n_decoders_cur = params.beam_search.beam_size;}} break;};n_decoders_cur = std::max(1, n_decoders_cur);WHISPER_PRINT_DEBUG("\n%s: strategy = %d, decoding with %d decoders, temperature = %.2f\n", __func__, params.strategy, n_decoders_cur, t_cur);// TAGS: WHISPER_DECODER_INITfor (int j = 0; j < n_decoders_cur; ++j) {auto & decoder = state->decoders[j];decoder.sequence.tokens.clear();decoder.sequence.result_len       = 0;decoder.sequence.sum_logprobs_all = 0.0;decoder.sequence.sum_logprobs     = -INFINITY;decoder.sequence.avg_logprobs     = -INFINITY;decoder.sequence.entropy          = 0.0;decoder.sequence.score            = -INFINITY;decoder.seek_delta = 100*WHISPER_CHUNK_SIZE;decoder.failed    = false;decoder.completed = false;decoder.has_ts    = false;if (params.grammar_rules != nullptr) {decoder.grammar = whisper_grammar_init(params.grammar_rules, params.n_grammar_rules, params.i_start_rule);} else {decoder.grammar = {};}}// init prompt and kv cache for the current iteration// TODO: do not recompute the prompt if it is the same as previous time{prompt.clear();// if we have already generated some text, use it as a prompt to condition the next generationif (!prompt_past.empty() && t_cur < 0.5f && params.n_max_text_ctx > 0) {int n_take = std::min(std::min(params.n_max_text_ctx, whisper_n_text_ctx(ctx)/2), int(prompt_past.size()));prompt = { whisper_token_prev(ctx) };prompt.insert(prompt.begin() + 1, prompt_past.end() - n_take, prompt_past.end());}// init new transcription with sot, language (opt) and task tokensprompt.insert(prompt.end(), prompt_init.begin(), prompt_init.end());// print the promptWHISPER_PRINT_DEBUG("\n\n");for (int i = 0; i < (int) prompt.size(); i++) {WHISPER_PRINT_DEBUG("%s: prompt[%d] = %s\n", __func__, i, ctx->vocab.id_to_token.at(prompt[i]).c_str());}WHISPER_PRINT_DEBUG("\n\n");whisper_kv_cache_clear(state->kv_self);whisper_batch_prep_legacy(state->batch, prompt.data(), prompt.size(), 0, 0);if (!whisper_decode_internal(*ctx, *state, state->batch, params.n_threads, params.abort_callback, params.abort_callback_user_data)) {WHISPER_LOG_ERROR("%s: failed to decode\n", __func__);return -7;}{const int64_t t_start_sample_us = ggml_time_us();state->decoders[0].i_batch = prompt.size() - 1;whisper_process_logits(*ctx, *state, state->decoders[0], params, t_cur);for (int j = 1; j < n_decoders_cur; ++j) {auto & decoder = state->decoders[j];whisper_kv_cache_seq_cp(state->kv_self, 0, j, -1, -1);memcpy(decoder.probs.data(),    state->decoders[0].probs.data(),    decoder.probs.size()*sizeof(decoder.probs[0]));memcpy(decoder.logits.data(),   state->decoders[0].logits.data(),   decoder.logits.size()*sizeof(decoder.logits[0]));memcpy(decoder.logprobs.data(), state->decoders[0].logprobs.data(), decoder.logprobs.size()*sizeof(decoder.logprobs[0]));}state->t_sample_us += ggml_time_us() - t_start_sample_us;}}for (int i = 0, n_max = whisper_n_text_ctx(ctx)/2 - 4; i < n_max; ++i) {const int64_t t_start_sample_us = ggml_time_us();if (params.strategy == whisper_sampling_strategy::WHISPER_SAMPLING_BEAM_SEARCH) {for (auto & bc : bc_per_dec) {bc.clear();}}// sampling// TODO: avoid memory allocations, optimize, avoid threads?{std::atomic<int> j_cur(0);auto process = [&]() {while (true) {const int j = j_cur.fetch_add(1);if (j >= n_decoders_cur) {break;}auto & decoder = state->decoders[j];if (decoder.completed || decoder.failed) {continue;}switch (params.strategy) {case whisper_sampling_strategy::WHISPER_SAMPLING_GREEDY:{if (t_cur < 1e-6f) {decoder.sequence.tokens.push_back(whisper_sample_token(*ctx, decoder, true));} else {decoder.sequence.tokens.push_back(whisper_sample_token(*ctx, decoder, false));}decoder.sequence.sum_logprobs_all += decoder.sequence.tokens.back().plog;} break;case whisper_sampling_strategy::WHISPER_SAMPLING_BEAM_SEARCH:{const auto tokens_new = whisper_sample_token_topk(*ctx, decoder, params.beam_search.beam_size);for (const auto & token : tokens_new) {bc_per_dec[j].push_back({ j, decoder.seek_delta, decoder.has_ts, decoder.sequence, decoder.grammar, });bc_per_dec[j].back().sequence.tokens.push_back(token);bc_per_dec[j].back().sequence.sum_logprobs_all += token.plog;}} break;};}};const int n_threads = std::min(params.n_threads, n_decoders_cur);if (n_threads == 1) {process();} else {std::vector<std::thread> threads(n_threads - 1);for (int t = 0; t < n_threads - 1; ++t) {threads[t] = std::thread(process);}process();for (int t = 0; t < n_threads - 1; ++t) {threads[t].join();}}}beam_candidates.clear();for (const auto & bc : bc_per_dec) {beam_candidates.insert(beam_candidates.end(), bc.begin(), bc.end());if (!bc.empty()) {state->n_sample += 1;}}// for beam-search, choose the top candidates and update the KV cachesif (params.strategy == whisper_sampling_strategy::WHISPER_SAMPLING_BEAM_SEARCH) {std::sort(beam_candidates.begin(),beam_candidates.end(),[](const beam_candidate & a, const beam_candidate & b) {return a.sequence.sum_logprobs_all > b.sequence.sum_logprobs_all;});uint32_t cur_c = 0;for (int j = 0; j < n_decoders_cur; ++j) {auto & decoder = state->decoders[j];if (decoder.completed || decoder.failed) {continue;}if (cur_c >= beam_candidates.size()) {cur_c = 0;}auto & cur = beam_candidates[cur_c++];while (beam_candidates.size() > cur_c && beam_candidates[cur_c].sequence.sum_logprobs_all == cur.sequence.sum_logprobs_all && i > 0) {++cur_c;}decoder.seek_delta = cur.seek_delta;decoder.has_ts     = cur.has_ts;decoder.sequence   = cur.sequence;decoder.grammar    = cur.grammar;whisper_kv_cache_seq_cp(state->kv_self, cur.decoder_idx, WHISPER_MAX_DECODERS + j, -1, -1);WHISPER_PRINT_DEBUG("%s: beam search: decoder %d: from decoder %d: token = %10s, plog = %8.5f, sum_logprobs = %8.5f\n",__func__, j, cur.decoder_idx, ctx->vocab.id_to_token.at(decoder.sequence.tokens.back().id).c_str(), decoder.sequence.tokens.back().plog, decoder.sequence.sum_logprobs_all);}for (int j = 0; j < n_decoders_cur; ++j) {auto & decoder = state->decoders[j];if (decoder.completed || decoder.failed) {continue;}whisper_kv_cache_seq_rm(state->kv_self, j,                           -1, -1);whisper_kv_cache_seq_cp(state->kv_self, WHISPER_MAX_DECODERS + j, j, -1, -1);whisper_kv_cache_seq_rm(state->kv_self, WHISPER_MAX_DECODERS + j,    -1, -1);}}// update the decoder state// - check if the sequence is completed// - check if the sequence is failed// - update sliding window based on timestamp tokensfor (int j = 0; j < n_decoders_cur; ++j) {auto & decoder = state->decoders[j];if (decoder.completed || decoder.failed) {continue;}auto & has_ts     = decoder.has_ts;auto & failed     = decoder.failed;auto & completed  = decoder.completed;auto & seek_delta = decoder.seek_delta;auto & result_len = decoder.sequence.result_len;{const auto & token = decoder.sequence.tokens.back();// timestamp token - update sliding windowif (token.id > whisper_token_beg(ctx)) {const int seek_delta_new = 2*(token.id - whisper_token_beg(ctx));// do not allow to go back in timeif (has_ts && seek_delta > seek_delta_new && result_len < i) {failed = true; // TODO: maybe this is not a failure ?continue;}seek_delta = seek_delta_new;result_len = i + 1;has_ts = true;}whisper_grammar_accept_token(*ctx, decoder.grammar, token.id);#ifdef WHISPER_DEBUG{const auto tt = token.pt > 0.10 ? ctx->vocab.id_to_token.at(token.tid) : "[?]";WHISPER_PRINT_DEBUG("%s: id = %3d, decoder = %d, token = %6d, p = %6.3f, ts = %10s, %6.3f, result_len = %4d '%s'\n",__func__, i, j, token.id, token.p, tt.c_str(), token.pt, result_len, ctx->vocab.id_to_token.at(token.id).c_str());}
#endif// end of segmentif (token.id == whisper_token_eot(ctx) ||               // end of text token(params.max_tokens > 0 && i >= params.max_tokens) || // max tokens per segment reached(has_ts && seek + seek_delta + 100 >= seek_end)      // end of audio reached) {if (result_len == 0) {if (seek + seek_delta + 100 >= seek_end) {result_len = i + 1;} else {failed = true;continue;}}if (params.single_segment) {result_len = i + 1;seek_delta = 100*WHISPER_CHUNK_SIZE;}completed = true;continue;}// TESTS: if no tensors are loaded, it means we are running testsif (ctx->model.n_loaded == 0) {seek_delta = 100*WHISPER_CHUNK_SIZE;completed = true;continue;}}// sometimes, the decoding can get stuck in a repetition loop// this is an attempt to mitigate such cases - we flag the decoding as failed and use a fallback strategyif (i == n_max - 1 && (result_len == 0 || seek_delta < 100*WHISPER_CHUNK_SIZE/2)) {failed = true;continue;}}// check if all decoders have finished (i.e. completed or failed){bool completed_all = true;for (int j = 0; j < n_decoders_cur; ++j) {auto & decoder = state->decoders[j];if (decoder.completed || decoder.failed) {continue;}completed_all = false;}if (completed_all) {break;}}state->t_sample_us += ggml_time_us() - t_start_sample_us;// obtain logits for the next token{auto & batch = state->batch;batch.n_tokens = 0;const int n_past = prompt.size() + i;for (int j = 0; j < n_decoders_cur; ++j) {auto & decoder = state->decoders[j];if (decoder.failed || decoder.completed) {continue;}//WHISPER_PRINT_DEBUG("%s: decoder %d: token %d, seek_delta %d\n", __func__, j, decoder.sequence.tokens.back().id, decoder.seek_delta);decoder.i_batch = batch.n_tokens;batch.token   [batch.n_tokens]    = decoder.sequence.tokens.back().id;batch.pos     [batch.n_tokens]    = n_past;batch.n_seq_id[batch.n_tokens]    = 1;batch.seq_id  [batch.n_tokens][0] = j;batch.logits  [batch.n_tokens]    = 1;batch.n_tokens++;}assert(batch.n_tokens > 0);if (!whisper_decode_internal(*ctx, *state, state->batch, params.n_threads, params.abort_callback, params.abort_callback_user_data)) {WHISPER_LOG_ERROR("%s: failed to decode\n", __func__);return -8;}const int64_t t_start_sample_us = ggml_time_us();// TODO: avoid memory allocations, optimize, avoid threads?{std::atomic<int> j_cur(0);auto process = [&]() {while (true) {const int j = j_cur.fetch_add(1);if (j >= n_decoders_cur) {break;}auto & decoder = state->decoders[j];if (decoder.failed || decoder.completed) {continue;}whisper_process_logits(*ctx, *state, decoder, params, t_cur);}};const int n_threads = std::min(params.n_threads, n_decoders_cur);if (n_threads == 1) {process();} else {std::vector<std::thread> threads(n_threads - 1);for (int t = 0; t < n_threads - 1; ++t) {threads[t] = std::thread(process);}process();for (int t = 0; t < n_threads - 1; ++t) {threads[t].join();}}}state->t_sample_us += ggml_time_us() - t_start_sample_us;}}// rank the resulting sequences and select the best one{double best_score = -INFINITY;for (int j = 0; j < n_decoders_cur; ++j) {auto & decoder = state->decoders[j];if (decoder.failed) {continue;}decoder.sequence.tokens.resize(decoder.sequence.result_len);whisper_sequence_score(params, decoder.sequence);WHISPER_PRINT_DEBUG("%s: decoder %2d: score = %8.5f, result_len = %3d, avg_logprobs = %8.5f, entropy = %8.5f\n",__func__, j, decoder.sequence.score, decoder.sequence.result_len, decoder.sequence.avg_logprobs, decoder.sequence.entropy);if (decoder.sequence.result_len > 32 && decoder.sequence.entropy < params.entropy_thold) {WHISPER_PRINT_DEBUG("%s: decoder %2d: failed due to entropy %8.5f < %8.5f\n",__func__, j, decoder.sequence.entropy, params.entropy_thold);decoder.failed = true;state->n_fail_h++;continue;}if (best_score < decoder.sequence.score) {best_score = decoder.sequence.score;best_decoder_id = j;}}WHISPER_PRINT_DEBUG("%s: best decoder = %d\n", __func__, best_decoder_id);}// was the decoding successful for the current temperature?// do fallback only if:// - we are not at the last temperature// - we are not at the end of the audio (3 sec)if (it != (int) temperatures.size() - 1 &&seek_end - seek > 10*WHISPER_CHUNK_SIZE) {bool success = true;const auto & decoder = state->decoders[best_decoder_id];if (decoder.failed || decoder.sequence.avg_logprobs < params.logprob_thold) {success = false;state->n_fail_p++;}if (success) {//for (auto & token : ctx->decoders[best_decoder_id].sequence.tokens) {//    WHISPER_PRINT_DEBUG("%s: token = %d, p = %6.3f, pt = %6.3f, ts = %s, str = %s\n", __func__, token.id, token.p, token.pt, ctx->vocab.id_to_token.at(token.tid).c_str(), ctx->vocab.id_to_token.at(token.id).c_str());//}break;}}WHISPER_PRINT_DEBUG("\n%s: failed to decode with temperature = %.2f\n", __func__, t_cur);}// output results through a user-provided callback{const auto & best_decoder = state->decoders[best_decoder_id];const auto seek_delta = best_decoder.seek_delta;const auto result_len = best_decoder.sequence.result_len;const auto & tokens_cur = best_decoder.sequence.tokens;//WHISPER_PRINT_DEBUG("prompt_init.size() = %d, prompt.size() = %d, result_len = %d, seek_delta = %d\n", prompt_init.size(), prompt.size(), result_len, seek_delta);// update prompt_pastprompt_past.clear();if (prompt.front() == whisper_token_prev(ctx)) {prompt_past.insert(prompt_past.end(), prompt.begin() + 1, prompt.end() - prompt_init.size());}for (int i = 0; i < result_len; ++i) {prompt_past.push_back(tokens_cur[i].id);}if (!tokens_cur.empty() && ctx->model.n_loaded > 0) {int  i0 = 0;auto t0 = seek + 2*(tokens_cur.front().tid - whisper_token_beg(ctx));std::string text;bool speaker_turn_next = false;for (int i = 0; i < (int) tokens_cur.size(); i++) {//printf("%s: %18s %6.3f %18s %6.3f\n", __func__,//        ctx->vocab.id_to_token[tokens_cur[i].id].c_str(), tokens_cur[i].p,//        ctx->vocab.id_to_token[tokens_cur[i].tid].c_str(), tokens_cur[i].pt);if (params.print_special || tokens_cur[i].id < whisper_token_eot(ctx)) {text += whisper_token_to_str(ctx, tokens_cur[i].id);}// [TDRZ] record if speaker turn was predicted after current segmentif (params.tdrz_enable && tokens_cur[i].id == whisper_token_solm(ctx)) {speaker_turn_next = true;}if (tokens_cur[i].id > whisper_token_beg(ctx) && !params.single_segment) {const auto t1 = seek + 2*(tokens_cur[i].tid - whisper_token_beg(ctx));if (!text.empty()) {const auto tt0 = params.speed_up ? 2*t0 : t0;const auto tt1 = params.speed_up ? 2*t1 : t1;if (params.print_realtime) {if (params.print_timestamps) {printf("[%s --> %s]  %s\n", to_timestamp(tt0).c_str(), to_timestamp(tt1).c_str(), text.c_str());} else {printf("%s", text.c_str());fflush(stdout);}}//printf("tt0 = %d, tt1 = %d, text = %s, token = %s, token_id = %d, tid = %d\n", tt0, tt1, text.c_str(), ctx->vocab.id_to_token[tokens_cur[i].id].c_str(), tokens_cur[i].id, tokens_cur[i].tid);result_all.push_back({ tt0, tt1, text, {}, speaker_turn_next });for (int j = i0; j <= i; j++) {result_all.back().tokens.push_back(tokens_cur[j]);}int n_new = 1;if (params.token_timestamps) {whisper_exp_compute_token_level_timestamps(*ctx, *state, result_all.size() - 1, params.thold_pt, params.thold_ptsum);if (params.max_len > 0) {n_new = whisper_wrap_segment(*ctx, *state, params.max_len, params.split_on_word);}}if (params.new_segment_callback) {params.new_segment_callback(ctx, state, n_new, params.new_segment_callback_user_data);}}text = "";while (i < (int) tokens_cur.size() && tokens_cur[i].id > whisper_token_beg(ctx)) {i++;}i--;t0 = t1;i0 = i + 1;speaker_turn_next = false;}}if (!text.empty()) {const auto t1 = seek + seek_delta;const auto tt0 = params.speed_up ? 2*t0 : t0;const auto tt1 = params.speed_up ? 2*t1 : t1;if (params.print_realtime) {if (params.print_timestamps) {printf("[%s --> %s]  %s\n", to_timestamp(tt0).c_str(), to_timestamp(tt1).c_str(), text.c_str());} else {printf("%s", text.c_str());fflush(stdout);}}result_all.push_back({ tt0, tt1, text, {} , speaker_turn_next });for (int j = i0; j < (int) tokens_cur.size(); j++) {result_all.back().tokens.push_back(tokens_cur[j]);}int n_new = 1;if (params.token_timestamps) {whisper_exp_compute_token_level_timestamps(*ctx, *state, result_all.size() - 1, params.thold_pt, params.thold_ptsum);if (params.max_len > 0) {n_new = whisper_wrap_segment(*ctx, *state, params.max_len, params.split_on_word);}}if (params.new_segment_callback) {params.new_segment_callback(ctx, state, n_new, params.new_segment_callback_user_data);}}}// update audio windowseek += seek_delta;WHISPER_PRINT_DEBUG("seek = %d, seek_delta = %d\n", seek, seek_delta);}}return 0;
}

3.5output stuff

        // output stuff{printf("\n");// output to text fileif (params.output_txt) {const auto fname_txt = fname_out + ".txt";output_txt(ctx, fname_txt.c_str(), params, pcmf32s);}// output to VTT fileif (params.output_vtt) {const auto fname_vtt = fname_out + ".vtt";output_vtt(ctx, fname_vtt.c_str(), params, pcmf32s);}// output to SRT fileif (params.output_srt) {const auto fname_srt = fname_out + ".srt";output_srt(ctx, fname_srt.c_str(), params, pcmf32s);}// output to WTS fileif (params.output_wts) {const auto fname_wts = fname_out + ".wts";output_wts(ctx, fname_wts.c_str(), fname_inp.c_str(), params, float(pcmf32.size() + 1000)/WHISPER_SAMPLE_RATE, pcmf32s);}// output to CSV fileif (params.output_csv) {const auto fname_csv = fname_out + ".csv";output_csv(ctx, fname_csv.c_str(), params, pcmf32s);}// output to JSON fileif (params.output_jsn) {const auto fname_jsn = fname_out + ".json";output_json(ctx, fname_jsn.c_str(), params, pcmf32s, params.output_jsn_full);}// output to LRC fileif (params.output_lrc) {const auto fname_lrc = fname_out + ".lrc";output_lrc(ctx, fname_lrc.c_str(), params, pcmf32s);}// output to score fileif (params.log_score) {const auto fname_score = fname_out + ".score.txt";output_score(ctx, fname_score.c_str(), params, pcmf32s);}}

stream

stream的依赖

if (WHISPER_SDL2) #  需要set(WHISPER_SDL2 ON)#option(WHISPER_SDL2 "whisper: support for libSDL2" OFF)# streamset(TARGET stream)add_executable(${TARGET} stream.cpp)include(DefaultTargetOptions)target_link_libraries(${TARGET} PRIVATE common common-sdl whisper ${CMAKE_THREAD_LIBS_INIT})
endif ()

// Real-time speech recognition of input from a microphone
//
// A very quick-n-dirty implementation serving mainly as a proof of concept.
//
#include "common-sdl.h" // https://github1s.com/ggerganov/whisper.cpp/blob/d6b9be21d76b91a96bb987063b25e5b532140253/examples/common-sdl.h
#include "common.h"
#include "whisper.h"#include <cassert>
#include <cstdio>
#include <string>
#include <thread>
#include <vector>
#include <fstream>

[ 66%] Linking CXX static library libcommon.a
gmake[2]: 离开目录“/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug”
[ 66%] Built target common
gmake[2]: 进入目录“/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug”
gmake[2]: 进入目录“/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug”
gmake[2]: 进入目录“/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug”
Scanning dependencies of target stream
gmake[2]: 离开目录“/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug”
gmake[2]: 离开目录“/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug”
gmake[2]: 进入目录“/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug”
gmake[2]: 进入目录“/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug”
gmake[2]: 离开目录“/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug”
[ 72%] Building CXX object examples/main/CMakeFiles/main.dir/main.cpp.o
gmake[2]: 进入目录“/home/pdd/le/whisper.cpp-1.5.0/cmake-build-debug”
[ 77%] Building CXX object examples/quantize/CMakeFiles/quantize.dir/quantize.cpp.o
[ 83%] Building CXX object examples/stream/CMakeFiles/stream.dir/stream.cpp.o
In file included from /home/pdd/le/whisper.cpp-1.5.0/examples/stream/stream.cpp:5:
/home/pdd/le/whisper.cpp-1.5.0/examples/common-sdl.h:3:10: fatal error: SDL.h: 没有那个文件或目录3 | #include <SDL.h>|          ^~~~~~~
compilation terminated.

SDL安装

        反正安装失败了，跟系统版本有关，各种依赖处理有点麻烦。

(base) pdd@pdd-Dell-G15-5511:~/le$ sudo apt install libsdl2-dev
[sudo] pdd 的密码： 
正在读取软件包列表... 完成
正在分析软件包的依赖关系树... 完成
正在读取状态信息... 完成                 
有一些软件包无法被安装。如果您用的是 unstable 发行版，这也许是
因为系统无法达到您要求的状态造成的。该版本中可能会有一些您需要的软件
包尚未被创建或是它们已被从新到(Incoming)目录移出。
下列信息可能会对解决问题有所帮助：下列软件包有未满足的依赖关系：udev : 破坏: systemd (< 249.11-0ubuntu3.11)破坏: systemd:i386 (< 249.11-0ubuntu3.11)推荐: systemd-hwe-hwdb 但是它将不会被安装
E: 错误，pkgProblemResolver::Resolve 发生故障，这可能是有软件包被要求保持现状的缘故。
(base) pdd@pdd-Dell-G15-5511:~/le$ sudo apt-get install libsdl2-dev
正在读取软件包列表... 完成
正在分析软件包的依赖关系树... 完成
正在读取状态信息... 完成                 
有一些软件包无法被安装。如果您用的是 unstable 发行版，这也许是
因为系统无法达到您要求的状态造成的。该版本中可能会有一些您需要的软件
包尚未被创建或是它们已被从新到(Incoming)目录移出。
下列信息可能会对解决问题有所帮助：下列软件包有未满足的依赖关系：udev : 破坏: systemd (< 249.11-0ubuntu3.11)破坏: systemd:i386 (< 249.11-0ubuntu3.11)推荐: systemd-hwe-hwdb 但是它将不会被安装
E: 错误，pkgProblemResolver::Resolve 发生故障，这可能是有软件包被要求保持现状的缘故。

(base) pdd@pdd-Dell-G15-5511:~/le$ sudo apt-get install libsdl2-2.0-0
正在读取软件包列表... 完成
正在分析软件包的依赖关系树... 完成
正在读取状态信息... 完成                 
下列软件包是自动安装的并且现在不需要了：fcitx-config-common fcitx-config-gtk fcitx-frontend-all fcitx-frontend-gtk2 fcitx-frontend-gtk3 fcitx-frontend-qt5 fcitx-module-dbusfcitx-module-kimpanel fcitx-module-lua fcitx-module-quickphrase-editor5 fcitx-module-x11 fcitx-modules fcitx-ui-classic g++-11 gir1.2-appindicator3-0.1gir1.2-gst-plugins-base-1.0 gir1.2-gstreamer-1.0 gir1.2-keybinder-3.0 gir1.2-wnck-3.0 gnome-session-canberra libfcitx-config4 libfcitx-core0libfcitx-gclient1 libfcitx-qt5-1 libfcitx-qt5-data libfcitx-utils0 libgettextpo0 libkeybinder-3.0-0 libpresage-data libpresage1v5 libtinyxml2.6.2v5libwnck-3-0 libwnck-3-common presage python3-gi-cairo
使用'sudo apt autoremove'来卸载它(它们)。
将会同时安装下列软件：libsndio6.1
建议安装：sndiod
下列【新】软件包将被安装：libsdl2-2.0-0 libsndio6.1
升级了 0 个软件包，新安装了 2 个软件包，要卸载 0 个软件包，有 28 个软件包未被升级。
需要下载 366 kB 的归档。
解压缩后会消耗 1,227 kB 的额外空间。
您希望继续执行吗？ [Y/n] y
获取:1 http://dk.archive.ubuntu.com/ubuntu xenial/universe amd64 libsndio6.1 amd64 1.1.0-2 [23.2 kB]
获取:2 http://dk.archive.ubuntu.com/ubuntu xenial/universe amd64 libsdl2-2.0-0 amd64 2.0.4+dfsg1-2ubuntu2 [343 kB]
已下载 366 kB，耗时 4秒 (99.5 kB/s)     
正在选中未选择的软件包 libsndio6.1:amd64。
(正在读取数据库 ... 系统当前共安装有 285392 个文件和目录。)
准备解压 .../libsndio6.1_1.1.0-2_amd64.deb  ...
正在解压 libsndio6.1:amd64 (1.1.0-2) ...
正在选中未选择的软件包 libsdl2-2.0-0:amd64。
准备解压 .../libsdl2-2.0-0_2.0.4+dfsg1-2ubuntu2_amd64.deb  ...
正在解压 libsdl2-2.0-0:amd64 (2.0.4+dfsg1-2ubuntu2) ...
正在设置 libsndio6.1:amd64 (1.1.0-2) ...
正在设置 libsdl2-2.0-0:amd64 (2.0.4+dfsg1-2ubuntu2) ...
正在处理用于 libc-bin (2.35-0ubuntu3.1) 的触发器 ...
/sbin/ldconfig.real: /usr/local/cuda-11.4/targets/x86_64-linux/lib/libcudnn_ops_infer.so.8 is not a symbolic link

$ sudo aptitude install libsdl2-dev
下列“新”软件包将被安装。         libsdl2-dev{b} libsndio-dev{a} 
下列软件包将被“删除”：fcitx-config-common{u} fcitx-config-gtk{u} fcitx-frontend-all{u} fcitx-frontend-gtk2{u} fcitx-frontend-gtk3{u} fcitx-frontend-qt5{u} fcitx-module-dbus{u} fcitx-module-kimpanel{u} fcitx-module-lua{u} fcitx-module-quickphrase-editor5{u} fcitx-module-x11{u} fcitx-modules{u} fcitx-ui-classic{u} g++-11{u} gir1.2-appindicator3-0.1{u} gir1.2-gst-plugins-base-1.0{u} gir1.2-gstreamer-1.0{u} gir1.2-keybinder-3.0{u} gir1.2-wnck-3.0{u} gnome-session-canberra{u} libfcitx-config4{u} libfcitx-core0{u} libfcitx-gclient1{u} libfcitx-qt5-1{u} libfcitx-qt5-data{u} libfcitx-utils0{u} libgettextpo0{u} libkeybinder-3.0-0{u} libpresage-data{u} libpresage1v5{u} libtinyxml2.6.2v5{u} libwnck-3-0{u} libwnck-3-common{u} presage{u} python3-gi-cairo{u} 
0 个软件包被升级，新安装 2 个，35 个将被删除， 同时 28 个将不升级。
需要获取 627 kB 的存档。解包后将释放 49.2 MB。
下列软件包存在未满足的依赖关系：libsdl2-dev : 依赖: libasound2-dev 但它是不可安装的依赖: libdbus-1-dev 但它是不可安装的依赖: libgles2-mesa-dev 但它是不可安装的依赖: libmirclient-dev 但它是不可安装的依赖: libpulse-dev 但它是不可安装的依赖: libudev-dev 但它是不可安装的依赖: libxkbcommon-dev 但它是不可安装的依赖: libxss-dev 但它是不可安装的依赖: libxv-dev 但它是不可安装的依赖: libxxf86vm-dev 但它是不可安装的
下列动作将解决这些依赖关系：保持 下列软件包于其当前版本：
1)     libsdl2-dev [未安装的]     是否接受该解决方案？[Y/n/q/?] y
下列软件包将被“删除”：fcitx-config-common{u} fcitx-config-gtk{u} fcitx-frontend-all{u} fcitx-frontend-gtk2{u} fcitx-frontend-gtk3{u} fcitx-frontend-qt5{u} fcitx-module-dbus{u} fcitx-module-kimpanel{u} fcitx-module-lua{u} fcitx-module-quickphrase-editor5{u} fcitx-module-x11{u} fcitx-modules{u} fcitx-ui-classic{u} g++-11{u} gir1.2-appindicator3-0.1{u} gir1.2-gst-plugins-base-1.0{u} gir1.2-gstreamer-1.0{u} gir1.2-keybinder-3.0{u} gir1.2-wnck-3.0{u} gnome-session-canberra{u} libfcitx-config4{u} libfcitx-core0{u} libfcitx-gclient1{u} libfcitx-qt5-1{u} libfcitx-qt5-data{u} libfcitx-utils0{u} libgettextpo0{u} libkeybinder-3.0-0{u} libpresage-data{u} libpresage1v5{u} libtinyxml2.6.2v5{u} libwnck-3-0{u} libwnck-3-common{u} presage{u} python3-gi-cairo{u} 
0 个软件包被升级，新安装 0 个，35 个将被删除， 同时 28 个将不升级。
需要获取 0 B 的存档。解包后将释放 53.1 MB。
您要继续吗？[Y/n/?] y
(正在读取数据库 ... 系统当前共安装有 285406 个文件和目录。)
正在卸载 fcitx-config-gtk (0.4.10-3) ...
正在卸载 fcitx-config-common (0.4.10-3) ...
正在卸载 fcitx-frontend-all (1:4.2.9.8-5) ...
正在卸载 fcitx-frontend-gtk2 (1:4.2.9.8-5) ...
正在卸载 fcitx-frontend-gtk3 (1:4.2.9.8-5) ...
正在卸载 fcitx-frontend-qt5:amd64 (1.2.7-1.2build1) ...
正在卸载 fcitx-module-kimpanel (1:4.2.9.8-5) ...
正在卸载 fcitx-module-dbus (1:4.2.9.8-5) ...
正在卸载 fcitx-module-lua (1:4.2.9.8-5) ...
正在卸载 fcitx-module-quickphrase-editor5:amd64 (1.2.7-1.2build1) ...
正在卸载 fcitx-ui-classic (1:4.2.9.8-5) ...
正在卸载 fcitx-module-x11 (1:4.2.9.8-5) ...
正在卸载 fcitx-modules (1:4.2.9.8-5) ...
正在卸载 g++-11 (11.3.0-1ubuntu1~22.04) ...
正在卸载 gir1.2-appindicator3-0.1 (12.10.1+20.10.20200706.1-0ubuntu1) ...
正在卸载 gir1.2-gst-plugins-base-1.0:amd64 (1.20.1-1) ...
正在卸载 gir1.2-gstreamer-1.0:amd64 (1.20.3-0ubuntu1) ...
正在卸载 gir1.2-keybinder-3.0 (0.3.2-1.1) ...
正在卸载 gir1.2-wnck-3.0:amd64 (40.1-1) ...
正在卸载 gnome-session-canberra (0.30-10ubuntu1) ...
正在卸载 libfcitx-qt5-1:amd64 (1.2.7-1.2build1) ...
正在卸载 libfcitx-core0:amd64 (1:4.2.9.8-5) ...
正在卸载 libfcitx-config4:amd64 (1:4.2.9.8-5) ...
正在卸载 libfcitx-gclient1:amd64 (1:4.2.9.8-5) ...
正在卸载 libfcitx-qt5-data (1.2.7-1.2build1) ...
正在卸载 libfcitx-utils0:amd64 (1:4.2.9.8-5) ...
正在卸载 libgettextpo0:amd64 (0.21-4ubuntu4) ...
正在卸载 libkeybinder-3.0-0:amd64 (0.3.2-1.1) ...
正在卸载 presage (0.9.1-2.2ubuntu1) ...
正在卸载 libpresage1v5:amd64 (0.9.1-2.2ubuntu1) ...
正在卸载 libpresage-data (0.9.1-2.2ubuntu1) ...
正在卸载 libtinyxml2.6.2v5:amd64 (2.6.2-6) ...
正在卸载 libwnck-3-0:amd64 (40.1-1) ...
正在卸载 libwnck-3-common (40.1-1) ...
正在卸载 python3-gi-cairo (3.42.1-0ubuntu1) ...
正在处理用于 mate-menus (1.26.0-2ubuntu2) 的触发器 ...
正在处理用于 libgtk-3-0:amd64 (3.24.33-1ubuntu2) 的触发器 ...
正在处理用于 libgtk2.0-0:amd64 (2.24.33-2ubuntu2) 的触发器 ...
正在处理用于 libc-bin (2.35-0ubuntu3.1) 的触发器 ...
/sbin/ldconfig.real: /usr/local/cuda-11.4/targets/x86_64-linux/lib/libcudnn_ops_infer.so.8 is not a symbolic link正在处理用于 man-db (2.10.2-1) 的触发器 ...
正在处理用于 mailcap (3.70+nmu1ubuntu1) 的触发器 ...
正在处理用于 desktop-file-utils (0.26-1ubuntu3) 的触发器 ...

编译安装 https://wiki.libsdl.org/SDL2/Installation

make

git clone https://github.com/libsdl-org/SDL.git -b SDL2
cd SDL
mkdir build
cd build
../configure
make
sudo make install

cmake

git clone https://github.com/libsdl-org/SDL
cd SDL
mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
cmake --build . --config Release --parallel#CMake >= 3.15
sudo cmake --install . --config Release#CMake <= 3.14
sudo make install

~/mysdl/SDL2-2.28.5/build$ sudo cmake --install . --config Release [sudo] pdd 的密码： -- Installing: /usr/local/lib/libSDL2-2.0.so.0.2800.5 -- Installing: /usr/local/lib/libSDL2-2.0.so.0 -- Installing: /usr/local/lib/libSDL2-2.0.so -- Installing: /usr/local/lib/libSDL2main.a -- Installing: /usr/local/lib/libSDL2.a -- Installing: /usr/local/lib/libSDL2_test.a -- Installing: /usr/local/lib/cmake/SDL2/SDL2Targets.cmake -- Installing: /usr/local/lib/cmake/SDL2/SDL2Targets-release.cmake -- Installing: /usr/local/lib/cmake/SDL2/SDL2mainTargets.cmake -- Installing: /usr/local/lib/cmake/SDL2/SDL2mainTargets-release.cmake -- Installing: /usr/local/lib/cmake/SDL2/SDL2staticTargets.cmake -- Installing: /usr/local/lib/cmake/SDL2/SDL2staticTargets-release.cmake -- Installing: /usr/local/lib/cmake/SDL2/SDL2testTargets.cmake -- Installing: /usr/local/lib/cmake/SDL2/SDL2testTargets-release.cmake -- Installing: /usr/local/lib/cmake/SDL2/SDL2Config.cmake -- Installing: /usr/local/lib/cmake/SDL2/SDL2ConfigVersion.cmake -- Installing: /usr/local/lib/cmake/SDL2/sdlfind.cmake -- Installing: /usr/local/include/SDL2/SDL.h -- Installing: /usr/local/include/SDL2/SDL_assert.h -- Installing: /usr/local/include/SDL2/SDL_atomic.h -- Installing: /usr/local/include/SDL2/SDL_audio.h -- Installing: /usr/local/include/SDL2/SDL_bits.h -- Installing: /usr/local/include/SDL2/SDL_blendmode.h -- Installing: /usr/local/include/SDL2/SDL_clipboard.h -- Installing: /usr/local/include/SDL2/SDL_copying.h -- Installing: /usr/local/include/SDL2/SDL_cpuinfo.h -- Installing: /usr/local/include/SDL2/SDL_egl.h -- Installing: /usr/local/include/SDL2/SDL_endian.h -- Installing: /usr/local/include/SDL2/SDL_error.h -- Installing: /usr/local/include/SDL2/SDL_events.h -- Installing: /usr/local/include/SDL2/SDL_filesystem.h -- Installing: /usr/local/include/SDL2/SDL_gamecontroller.h -- Installing: /usr/local/include/SDL2/SDL_gesture.h -- Installing: /usr/local/include/SDL2/SDL_guid.h -- Installing: /usr/local/include/SDL2/SDL_haptic.h -- Installing: /usr/local/include/SDL2/SDL_hidapi.h -- Installing: /usr/local/include/SDL2/SDL_hints.h -- Installing: /usr/local/include/SDL2/SDL_joystick.h -- Installing: /usr/local/include/SDL2/SDL_keyboard.h -- Installing: /usr/local/include/SDL2/SDL_keycode.h -- Installing: /usr/local/include/SDL2/SDL_loadso.h -- Installing: /usr/local/include/SDL2/SDL_locale.h -- Installing: /usr/local/include/SDL2/SDL_log.h -- Installing: /usr/local/include/SDL2/SDL_main.h -- Installing: /usr/local/include/SDL2/SDL_messagebox.h -- Installing: /usr/local/include/SDL2/SDL_metal.h -- Installing: /usr/local/include/SDL2/SDL_misc.h -- Installing: /usr/local/include/SDL2/SDL_mouse.h -- Installing: /usr/local/include/SDL2/SDL_mutex.h -- Installing: /usr/local/include/SDL2/SDL_name.h -- Installing: /usr/local/include/SDL2/SDL_opengl.h -- Installing: /usr/local/include/SDL2/SDL_opengl_glext.h -- Installing: /usr/local/include/SDL2/SDL_opengles.h -- Installing: /usr/local/include/SDL2/SDL_opengles2.h -- Installing: /usr/local/include/SDL2/SDL_opengles2_gl2.h -- Installing: /usr/local/include/SDL2/SDL_opengles2_gl2ext.h -- Installing: /usr/local/include/SDL2/SDL_opengles2_gl2platform.h -- Installing: /usr/local/include/SDL2/SDL_opengles2_khrplatform.h -- Installing: /usr/local/include/SDL2/SDL_pixels.h -- Installing: /usr/local/include/SDL2/SDL_platform.h -- Installing: /usr/local/include/SDL2/SDL_power.h -- Installing: /usr/local/include/SDL2/SDL_quit.h -- Installing: /usr/local/include/SDL2/SDL_rect.h -- Installing: /usr/local/include/SDL2/SDL_render.h -- Installing: /usr/local/include/SDL2/SDL_rwops.h -- Installing: /usr/local/include/SDL2/SDL_scancode.h -- Installing: /usr/local/include/SDL2/SDL_sensor.h -- Installing: /usr/local/include/SDL2/SDL_shape.h -- Installing: /usr/local/include/SDL2/SDL_stdinc.h -- Installing: /usr/local/include/SDL2/SDL_surface.h -- Installing: /usr/local/include/SDL2/SDL_system.h -- Installing: /usr/local/include/SDL2/SDL_syswm.h -- Installing: /usr/local/include/SDL2/SDL_test.h -- Installing: /usr/local/include/SDL2/SDL_test_assert.h -- Installing: /usr/local/include/SDL2/SDL_test_common.h -- Installing: /usr/local/include/SDL2/SDL_test_compare.h -- Installing: /usr/local/include/SDL2/SDL_test_crc32.h -- Installing: /usr/local/include/SDL2/SDL_test_font.h -- Installing: /usr/local/include/SDL2/SDL_test_fuzzer.h -- Installing: /usr/local/include/SDL2/SDL_test_harness.h -- Installing: /usr/local/include/SDL2/SDL_test_images.h -- Installing: /usr/local/include/SDL2/SDL_test_log.h -- Installing: /usr/local/include/SDL2/SDL_test_md5.h -- Installing: /usr/local/include/SDL2/SDL_test_memory.h -- Installing: /usr/local/include/SDL2/SDL_test_random.h -- Installing: /usr/local/include/SDL2/SDL_thread.h -- Installing: /usr/local/include/SDL2/SDL_timer.h -- Installing: /usr/local/include/SDL2/SDL_touch.h -- Installing: /usr/local/include/SDL2/SDL_types.h -- Installing: /usr/local/include/SDL2/SDL_version.h -- Installing: /usr/local/include/SDL2/SDL_video.h -- Installing: /usr/local/include/SDL2/SDL_vulkan.h -- Installing: /usr/local/include/SDL2/begin_code.h -- Installing: /usr/local/include/SDL2/close_code.h -- Installing: /usr/local/include/SDL2/SDL_revision.h -- Installing: /usr/local/include/SDL2/SDL_config.h -- Installing: /usr/local/share/licenses/SDL2/LICENSE.txt -- Installing: /usr/local/lib/pkgconfig/sdl2.pc -- Installing: /usr/local/lib/libSDL2.so -- Installing: /usr/local/bin/sdl2-config -- Installing: /usr/local/share/aclocal/sdl2.m4

• ERROR: Couldn’t initialize SDL: dsp: No such audio device
  

CG

static const std::map<std::string, std::pair<int, std::string>> g_lang = {{ "en",  { 0,  "english",         } },{ "zh",  { 1,  "chinese",         } },{ "de",  { 2,  "german",          } },{ "es",  { 3,  "spanish",         } },{ "ru",  { 4,  "russian",         } },{ "ko",  { 5,  "korean",          } },{ "fr",  { 6,  "french",          } },{ "ja",  { 7,  "japanese",        } },{ "pt",  { 8,  "portuguese",      } },{ "tr",  { 9,  "turkish",         } },{ "pl",  { 10, "polish",          } },{ "ca",  { 11,  "catalan",        } },{ "nl",  { 12,  "dutch",          } },{ "ar",  { 13,  "arabic",         } },{ "sv",  { 14,  "swedish",        } },{ "it",  { 15,  "italian",        } },{ "id",  { 16,  "indonesian",     } },{ "hi",  { 17,  "hindi",          } },{ "fi",  { 18,  "finnish",        } },{ "vi",  { 19,  "vietnamese",     } },{ "he",  { 20,  "hebrew",         } },{ "uk",  { 21,  "ukrainian",      } },{ "el",  { 22,  "greek",          } },{ "ms",  { 23,  "malay",          } },{ "cs",  { 24,  "czech",          } },{ "ro",  { 25,  "romanian",       } },{ "da",  { 26,  "danish",         } },{ "hu",  { 27,  "hungarian",      } },{ "ta",  { 28,  "tamil",          } },{ "no",  { 29,  "norwegian",      } },{ "th",  { 30,  "thai",           } },{ "ur",  { 31,  "urdu",           } },{ "hr",  { 32,  "croatian",       } },{ "bg",  { 33,  "bulgarian",      } },{ "lt",  { 34,  "lithuanian",     } },{ "la",  { 35,  "latin",          } },{ "mi",  { 36,  "maori",          } },{ "ml",  { 37,  "malayalam",      } },{ "cy",  { 38,  "welsh",          } },{ "sk",  { 39,  "slovak",         } },{ "te",  { 40,  "telugu",         } },{ "fa",  { 41,  "persian",        } },{ "lv",  { 42,  "latvian",        } },{ "bn",  { 43,  "bengali",        } },{ "sr",  { 44,  "serbian",        } },{ "az",  { 45,  "azerbaijani",    } },{ "sl",  { 46,  "slovenian",      } },{ "kn",  { 47,  "kannada",        } },{ "et",  { 48,  "estonian",       } },{ "mk",  { 49,  "macedonian",     } },{ "br",  { 50,  "breton",         } },{ "eu",  { 51,  "basque",         } },{ "is",  { 52,  "icelandic",      } },{ "hy",  { 53,  "armenian",       } },{ "ne",  { 54,  "nepali",         } },{ "mn",  { 55,  "mongolian",      } },{ "bs",  { 56,  "bosnian",        } },{ "kk",  { 57,  "kazakh",         } },{ "sq",  { 58,  "albanian",       } },{ "sw",  { 59,  "swahili",        } },{ "gl",  { 60,  "galician",       } },{ "mr",  { 61,  "marathi",        } },{ "pa",  { 62,  "punjabi",        } },{ "si",  { 63,  "sinhala",        } },{ "km",  { 64,  "khmer",          } },{ "sn",  { 65,  "shona",          } },{ "yo",  { 66,  "yoruba",         } },{ "so",  { 67,  "somali",         } },{ "af",  { 68,  "afrikaans",      } },{ "oc",  { 69,  "occitan",        } },{ "ka",  { 70,  "georgian",       } },{ "be",  { 71,  "belarusian",     } },{ "tg",  { 72,  "tajik",          } },{ "sd",  { 73,  "sindhi",         } },{ "gu",  { 74,  "gujarati",       } },{ "am",  { 75,  "amharic",        } },{ "yi",  { 76,  "yiddish",        } },{ "lo",  { 77,  "lao",            } },{ "uz",  { 78,  "uzbek",          } },{ "fo",  { 79,  "faroese",        } },{ "ht",  { 80,  "haitian creole", } },{ "ps",  { 81,  "pashto",         } },{ "tk",  { 82,  "turkmen",        } },{ "nn",  { 83,  "nynorsk",        } },{ "mt",  { 84,  "maltese",        } },{ "sa",  { 85,  "sanskrit",       } },{ "lb",  { 86,  "luxembourgish",  } },{ "my",  { 87,  "myanmar",        } },{ "bo",  { 88,  "tibetan",        } },{ "tl",  { 89,  "tagalog",        } },{ "mg",  { 90,  "malagasy",       } },{ "as",  { 91,  "assamese",       } },{ "tt",  { 92,  "tatar",          } },{ "haw", { 93,  "hawaiian",       } },{ "ln",  { 94,  "lingala",        } },{ "ha",  { 95,  "hausa",          } },{ "ba",  { 96,  "bashkir",        } },{ "jw",  { 97,  "javanese",       } },{ "su",  { 98,  "sundanese",      } },{ "yue", { 99,  "cantonese",      } },
};

(base) pdd@pdd-Dell-G15-5511:~/le$ git clone http://github.com/hogelog/whispercppapp.git --recurse-submodules
正克隆到 'whispercppapp'...
warning: 重定向到 https://github.com/hogelog/whispercppapp.git/
remote: Enumerating objects: 411, done.
remote: Counting objects: 100% (411/411), done.
remote: Compressing objects: 100% (245/245), done.
remote: Total 411 (delta 150), reused 368 (delta 113), pack-reused 0
接收对象中: 100% (411/411), 454.29 KiB | 161.00 KiB/s, 完成.
处理 delta 中: 100% (150/150), 完成.
子模组 'whisper.cpp'（https://github.com/ggerganov/whisper.cpp.git）已对路径 'whisper.cpp' 注册
正克隆到 '/home/pdd/le/whispercppapp/whisper.cpp'...
remote: Enumerating objects: 6590, done.        
remote: Counting objects: 100% (1812/1812), done.        
remote: Compressing objects: 100% (192/192), done.        
error: RPC 失败。curl 16 Error in the HTTP2 framing layer
error: 预期仍然需要 5253 个字节的正文
fetch-pack: unexpected disconnect while reading sideband packet
fatal: 过早的文件结束符（EOF）
fatal: fetch-pack：无效的 index-pack 输出
fatal: 无法克隆 'https://github.com/ggerganov/whisper.cpp.git' 到子模组路径 '/home/pdd/le/whispercppapp/whisper.cpp'
克隆 'whisper.cpp' 失败。按计划重试
正克隆到 '/home/pdd/le/whispercppapp/whisper.cpp'...
remote: Enumerating objects: 6590, done.        
remote: Counting objects: 100% (1807/1807), done.        
remote: Compressing objects: 100% (191/191), done.        
remote: Total 6590 (delta 1699), reused 1651 (delta 1613), pack-reused 4783        
接收对象中: 100% (6590/6590), 9.99 MiB | 209.00 KiB/s, 完成.
处理 delta 中: 100% (4244/4244), 完成.
子模组路径 'whisper.cpp'：检出 'ad1389003d3f8bd47b8ca7d4c21b4764cc3844fc'
子模组 'bindings/ios'（https://github.com/ggerganov/whisper.spm）已对路径 'whisper.cpp/bindings/ios' 注册
正克隆到 '/home/pdd/le/whispercppapp/whisper.cpp/bindings/ios'...
remote: Enumerating objects: 357, done.        
remote: Counting objects: 100% (151/151), done.        
remote: Compressing objects: 100% (71/71), done.        
remote: Total 357 (delta 104), reused 104 (delta 80), pack-reused 206        
接收对象中: 100% (357/357), 1.11 MiB | 163.00 KiB/s, 完成.
处理 delta 中: 100% (197/197), 完成.
子模组路径 'whisper.cpp/bindings/ios'：检出 '92d4c5c9a07b726e35c20dc513532789919e00c4'
子模组路径 'whisper.cpp'：检出 'ad1389003d3f8bd47b8ca7d4c21b4764cc3844fc'