Skip to content

Commit 1cc41dd

Browse files
data-mandata-man
committed
Use miniaudio for direct decoding flac, mp3, ogg and wav
1 parent d682e15 commit 1cc41dd

File tree

9 files changed

+98985
-8912
lines changed

9 files changed

+98985
-8912
lines changed

‎Makefile

Lines changed: 1 addition & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -18,17 +18,6 @@ samples:
1818
@wget --quiet --show-progress -O samples/mm1.wav https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav
1919
@wget --quiet --show-progress -O samples/a13.mp3 https://upload.wikimedia.org/wikipedia/commons/transcoded/6/6f/Apollo13-wehaveaproblem.ogg/Apollo13-wehaveaproblem.ogg.mp3
2020
@wget --quiet --show-progress -O samples/diffusion2023-07-03.flac https://archive.org/download/diffusion2023-07-03/diffusion2023-07-03.flac
21-
@echo "Converting to 16-bit WAV ..."
22-
@ffmpeg -loglevel -0 -y -i samples/gb0.ogg -ar 16000 -ac 1 -c:a pcm_s16le samples/gb0.wav
23-
@ffmpeg -loglevel -0 -y -i samples/gb1.ogg -ar 16000 -ac 1 -c:a pcm_s16le samples/gb1.wav
24-
@ffmpeg -loglevel -0 -y -i samples/hp0.ogg -ar 16000 -ac 1 -c:a pcm_s16le samples/hp0.wav
25-
@rm samples/*.ogg
26-
@ffmpeg -loglevel -0 -y -i samples/mm1.wav -ar 16000 -ac 1 -c:a pcm_s16le samples/mm0.wav
27-
@rm samples/mm1.wav
28-
@ffmpeg -loglevel -0 -y -i samples/a13.mp3 -ar 16000 -ac 1 -c:a pcm_s16le -ss 00:00:00 -to 00:00:30 samples/a13.wav
29-
@rm samples/a13.mp3
30-
@ffmpeg -loglevel -0 -y -i samples/diffusion2023-07-03.flac -ar 16000 -ac 1 -c:a pcm_s16le samples/diffusion2023-07-03.wav
31-
@rm samples/diffusion2023-07-03.flac
3221

3322
#
3423
# Models
@@ -59,7 +48,7 @@ tiny.en tiny base.en base small.en small medium.en medium large-v1 large-v2 larg
5948
@echo "Running $@ on all samples in ./samples ..."
6049
@echo "==============================================="
6150
@echo ""
62-
@for f in samples/*.wav; do \
51+
@for f in samples/*$(.flac .mp3 .ogg .wav); do \
6352
echo "----------------------------------------------" ; \
6453
echo "[+] Running $@ on $$f ... (run 'ffplay $$f' to listen)" ; \
6554
echo "----------------------------------------------" ; \

‎examples/cli/cli.cpp

Lines changed: 5 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -199,7 +199,8 @@ static bool whisper_params_parse(int argc, char ** argv, whisper_params & params
199199

200200
static void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & params) {
201201
fprintf(stderr, "\n");
202-
fprintf(stderr, "usage: %s [options] file0.wav file1.wav ...\n", argv[0]);
202+
fprintf(stderr, "usage: %s [options] file0 file1 ...\n", argv[0]);
203+
fprintf(stderr, "supported audio formats: flac, mp3, ogg, wav\n");
203204
fprintf(stderr, "\n");
204205
fprintf(stderr, "options:\n");
205206
fprintf(stderr, " -h, --help [default] show this help message and exit\n");
@@ -244,7 +245,7 @@ static void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params
244245
fprintf(stderr, " -dl, --detect-language [%-7s] exit after automatically detecting language\n", params.detect_language ? "true" : "false");
245246
fprintf(stderr, " --prompt PROMPT [%-7s] initial prompt (max n_text_ctx/2 tokens)\n", params.prompt.c_str());
246247
fprintf(stderr, " -m FNAME, --model FNAME [%-7s] model path\n", params.model.c_str());
247-
fprintf(stderr, " -f FNAME, --file FNAME [%-7s] input WAV file path\n", "");
248+
fprintf(stderr, " -f FNAME, --file FNAME [%-7s] input audio file path\n", "");
248249
fprintf(stderr, " -oved D, --ov-e-device DNAME [%-7s] the OpenVINO device used for encode inference\n", params.openvino_encode_device.c_str());
249250
fprintf(stderr, " -dtw MODEL --dtw MODEL [%-7s] compute token-level timestamps\n", params.dtw.c_str());
250251
fprintf(stderr, " -ls, --log-score [%-7s] log best decoder scores of tokens\n", params.log_score?"true":"false");
@@ -1069,8 +1070,8 @@ int main(int argc, char ** argv) {
10691070
std::vector<float> pcmf32; // mono-channel F32 PCM
10701071
std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
10711072

1072-
if (!::read_wav(fname_inp, pcmf32, pcmf32s, params.diarize)) {
1073-
fprintf(stderr, "error: failed to read WAV file '%s'\n", fname_inp.c_str());
1073+
if (!::read_audio_data(fname_inp, pcmf32, pcmf32s, params.diarize)) {
1074+
fprintf(stderr, "error: failed to read audio file '%s'\n", fname_inp.c_str());
10741075
continue;
10751076
}
10761077

‎examples/common.cpp

Lines changed: 70 additions & 69 deletions
Original file line numberDiff line numberDiff line change
@@ -4,8 +4,17 @@
44

55
// third-party utilities
66
// use your favorite implementations
7-
#define DR_WAV_IMPLEMENTATION
8-
#include "dr_wav.h"
7+
#define STB_VORBIS_HEADER_ONLY
8+
#include "stb_vorbis.c" /* Enables Vorbis decoding. */
9+
10+
#define MA_NO_DEVICE_IO
11+
#define MA_NO_THREADING
12+
#define MA_NO_ENCODING
13+
#define MA_NO_GENERATION
14+
#define MA_NO_RESOURCE_MANAGER
15+
#define MA_NO_NODE_GRAPH
16+
#define MINIAUDIO_IMPLEMENTATION
17+
#include "miniaudio.h"
918

1019
#include <cmath>
1120
#include <cstring>
@@ -639,9 +648,14 @@ bool is_wav_buffer(const std::string buf) {
639648
return true;
640649
}
641650

642-
bool read_wav(const std::string & fname, std::vector<float>& pcmf32, std::vector<std::vector<float>>& pcmf32s, bool stereo) {
643-
drwav wav;
644-
std::vector<uint8_t> wav_data; // used for pipe input from stdin or ffmpeg decoding output
651+
bool read_audio_data(const std::string & fname, std::vector<float>& pcmf32, std::vector<std::vector<float>>& pcmf32s, bool stereo) {
652+
std::vector<uint8_t> audio_data; // used for pipe input from stdin or ffmpeg decoding output
653+
654+
ma_result result;
655+
ma_decoder_config decoder_config;
656+
ma_decoder decoder;
657+
658+
decoder_config = ma_decoder_config_init(ma_format_f32, stereo ? 2 : 1, COMMON_SAMPLE_RATE);
645659

646660
if (fname == "-") {
647661
{
@@ -656,94 +670,78 @@ bool read_wav(const std::string & fname, std::vector<float>& pcmf32, std::vector
656670
if (n == 0) {
657671
break;
658672
}
659-
wav_data.insert(wav_data.end(), buf, buf + n);
673+
audio_data.insert(audio_data.end(), buf, buf + n);
660674
}
661675
}
662676

663-
if (drwav_init_memory(&wav, wav_data.data(), wav_data.size(), nullptr) == false) {
664-
fprintf(stderr, "error: failed to open WAV file from stdin\n");
665-
return false;
666-
}
677+
if ((result = ma_decoder_init_memory(audio_data.data(), audio_data.size(), &decoder_config, &decoder)) != MA_SUCCESS) {
678+
679+
fprintf(stderr, "Error: failed to open audio data from stdin (%s)\n", ma_result_description(result));
667680

668-
fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, wav_data.size());
681+
return false;
682+
}
683+
684+
fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, audio_data.size());
669685
}
670686
else if (is_wav_buffer(fname)) {
671-
if (drwav_init_memory(&wav, fname.c_str(), fname.size(), nullptr) == false) {
672-
fprintf(stderr, "error: failed to open WAV file from fname buffer\n");
673-
return false;
674-
}
687+
if ((result = ma_decoder_init_memory(audio_data.data(), audio_data.size(), &decoder_config, &decoder)) != MA_SUCCESS) {
688+
689+
fprintf(stderr, "Error: failed to open audio data from fname buffer (%s)\n", ma_result_description(result));
690+
691+
return false;
692+
}
675693
}
676-
else if (drwav_init_file(&wav, fname.c_str(), nullptr) == false) {
694+
else if ((result = ma_decoder_init_file(fname.c_str(), &decoder_config, &decoder)) != MA_SUCCESS) {
677695
#if defined(WHISPER_FFMPEG)
678-
if (ffmpeg_decode_audio(fname, wav_data) != 0) {
679-
fprintf(stderr, "error: failed to ffmpeg decode '%s' \n", fname.c_str());
680-
return false;
681-
}
682-
if (drwav_init_memory(&wav, wav_data.data(), wav_data.size(), nullptr) == false) {
683-
fprintf(stderr, "error: failed to read wav data as wav \n");
684-
return false;
685-
}
696+
if (ffmpeg_decode_audio(fname, audio_data) != 0) {
697+
698+
fprintf(stderr, "error: failed to ffmpeg decode '%s'\n", fname.c_str());
699+
700+
return false;
701+
}
702+
if ((result = ma_decoder_init_memory(audio_data.data(), audio_data.size(), &decoder_config, &decoder)) != MA_SUCCESS) {
703+
704+
fprintf(stderr, "error: failed to read audio data as wav (%s)\n", ma_result_description(result));
705+
706+
return false;
707+
}
686708
#else
687-
fprintf(stderr, "error: failed to open '%s' as WAV file\n", fname.c_str());
688-
return false;
709+
fprintf(stderr, "error: failed to open '%s' file (%s)\n", fname.c_str(), ma_result_description(result));
710+
711+
return false;
689712
#endif
690713
}
691714

692-
if (wav.channels != 1 && wav.channels != 2) {
693-
fprintf(stderr, "%s: WAV file '%s' must be mono or stereo\n", __func__, fname.c_str());
694-
drwav_uninit(&wav);
695-
return false;
696-
}
715+
ma_uint64 frame_count;
716+
ma_uint64 frames_read;
697717

698-
if (stereo && wav.channels != 2) {
699-
fprintf(stderr, "%s: WAV file '%s' must be stereo for diarization\n", __func__, fname.c_str());
700-
drwav_uninit(&wav);
701-
return false;
702-
}
718+
if ((result = ma_decoder_get_length_in_pcm_frames(&decoder, &frame_count)) != MA_SUCCESS) {
703719

704-
if (wav.sampleRate != COMMON_SAMPLE_RATE) {
705-
fprintf(stderr, "%s: WAV file '%s' must be %i kHz\n", __func__, fname.c_str(), COMMON_SAMPLE_RATE/1000);
706-
drwav_uninit(&wav);
707-
return false;
708-
}
720+
fprintf(stderr, "error: failed to retrieve the length of the audio data (%s)\n", ma_result_description(result));
709721

710-
if (wav.bitsPerSample != 16) {
711-
fprintf(stderr, "%s: WAV file '%s' must be 16-bit\n", __func__, fname.c_str());
712-
drwav_uninit(&wav);
713-
return false;
722+
return false;
714723
}
715724

716-
const uint64_t n = wav_data.empty() ? wav.totalPCMFrameCount : wav_data.size()/(wav.channels*wav.bitsPerSample/8);
725+
pcmf32.resize(stereo ? frame_count*2 : frame_count);
717726

718-
std::vector<int16_t> pcm16;
719-
pcm16.resize(n*wav.channels);
720-
drwav_read_pcm_frames_s16(&wav, n, pcm16.data());
721-
drwav_uninit(&wav);
727+
if ((result = ma_decoder_read_pcm_frames(&decoder, pcmf32.data(), frame_count, &frames_read)) != MA_SUCCESS) {
722728

723-
// convert to mono, float
724-
pcmf32.resize(n);
725-
if (wav.channels == 1) {
726-
for (uint64_t i = 0; i < n; i++) {
727-
pcmf32[i] = float(pcm16[i])/32768.0f;
728-
}
729-
} else {
730-
for (uint64_t i = 0; i < n; i++) {
731-
pcmf32[i] = float(pcm16[2*i] + pcm16[2*i + 1])/65536.0f;
732-
}
729+
fprintf(stderr, "error: failed to read the frames of the audio data (%s)\n", ma_result_description(result));
730+
731+
return false;
733732
}
734733

735734
if (stereo) {
736-
// convert to stereo, float
737-
pcmf32s.resize(2);
738-
739-
pcmf32s[0].resize(n);
740-
pcmf32s[1].resize(n);
741-
for (uint64_t i = 0; i < n; i++) {
742-
pcmf32s[0][i] = float(pcm16[2*i])/32768.0f;
743-
pcmf32s[1][i] = float(pcm16[2*i + 1])/32768.0f;
744-
}
735+
pcmf32s[0].resize(frame_count);
736+
pcmf32s[1].resize(frame_count);
737+
for (uint64_t i = 0; i < frame_count; i++) {
738+
pcmf32s[0][i] = pcmf32[2*i];
739+
pcmf32s[1][i] = pcmf32[2*i + 1];
740+
}
745741
}
746742

743+
ma_decoder_uninit(&decoder);
744+
747745
return true;
748746
}
749747

@@ -909,3 +907,6 @@ bool speak_with_file(const std::string & command, const std::string & text, cons
909907
}
910908
return true;
911909
}
910+
911+
#undef STB_VORBIS_HEADER_ONLY
912+
#include "stb_vorbis.c"

‎examples/common.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -143,7 +143,7 @@ bool is_wav_buffer(const std::string buf);
143143
// fname can be a buffer of WAV data instead of a filename
144144
// The sample rate of the audio must be equal to COMMON_SAMPLE_RATE
145145
// If stereo flag is set and the audio has 2 channels, the pcmf32s will contain 2 channel PCM
146-
bool read_wav(
146+
bool read_audio_data(
147147
const std::string & fname,
148148
std::vector<float> & pcmf32,
149149
std::vector<std::vector<float>> & pcmf32s,

0 commit comments

Comments
 (0)