whispercpp 1.3.1 → 1.3.2

data/ext/sources/examples/common-sdl.cpp

@@ -0,0 +1,227 @@
+#include "common-sdl.h"
+
+#include <cstdio>
+
+audio_async::audio_async(int len_ms) {
+    m_len_ms = len_ms;
+
+    m_running = false;
+}
+
+audio_async::~audio_async() {
+    if (m_dev_id_in) {
+        SDL_CloseAudioDevice(m_dev_id_in);
+    }
+}
+
+bool audio_async::init(int capture_id, int sample_rate) {
+    SDL_LogSetPriority(SDL_LOG_CATEGORY_APPLICATION, SDL_LOG_PRIORITY_INFO);
+
+    if (SDL_Init(SDL_INIT_AUDIO) < 0) {
+        SDL_LogError(SDL_LOG_CATEGORY_APPLICATION, "Couldn't initialize SDL: %s\n", SDL_GetError());
+        return false;
+    }
+
+    SDL_SetHintWithPriority(SDL_HINT_AUDIO_RESAMPLING_MODE, "medium", SDL_HINT_OVERRIDE);
+
+    {
+        int nDevices = SDL_GetNumAudioDevices(SDL_TRUE);
+        fprintf(stderr, "%s: found %d capture devices:\n", __func__, nDevices);
+        for (int i = 0; i < nDevices; i++) {
+            fprintf(stderr, "%s:    - Capture device #%d: '%s'\n", __func__, i, SDL_GetAudioDeviceName(i, SDL_TRUE));
+        }
+    }
+
+    SDL_AudioSpec capture_spec_requested;
+    SDL_AudioSpec capture_spec_obtained;
+
+    SDL_zero(capture_spec_requested);
+    SDL_zero(capture_spec_obtained);
+
+    capture_spec_requested.freq     = sample_rate;
+    capture_spec_requested.format   = AUDIO_F32;
+    capture_spec_requested.channels = 1;
+    capture_spec_requested.samples  = 1024;
+    capture_spec_requested.callback = [](void * userdata, uint8_t * stream, int len) {
+        audio_async * audio = (audio_async *) userdata;
+        audio->callback(stream, len);
+    };
+    capture_spec_requested.userdata = this;
+
+    if (capture_id >= 0) {
+        fprintf(stderr, "%s: attempt to open capture device %d : '%s' ...\n", __func__, capture_id, SDL_GetAudioDeviceName(capture_id, SDL_TRUE));
+        m_dev_id_in = SDL_OpenAudioDevice(SDL_GetAudioDeviceName(capture_id, SDL_TRUE), SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
+    } else {
+        fprintf(stderr, "%s: attempt to open default capture device ...\n", __func__);
+        m_dev_id_in = SDL_OpenAudioDevice(nullptr, SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
+    }
+
+    if (!m_dev_id_in) {
+        fprintf(stderr, "%s: couldn't open an audio device for capture: %s!\n", __func__, SDL_GetError());
+        m_dev_id_in = 0;
+
+        return false;
+    } else {
+        fprintf(stderr, "%s: obtained spec for input device (SDL Id = %d):\n", __func__, m_dev_id_in);
+        fprintf(stderr, "%s:     - sample rate:       %d\n",                   __func__, capture_spec_obtained.freq);
+        fprintf(stderr, "%s:     - format:            %d (required: %d)\n",    __func__, capture_spec_obtained.format,
+                capture_spec_requested.format);
+        fprintf(stderr, "%s:     - channels:          %d (required: %d)\n",    __func__, capture_spec_obtained.channels,
+                capture_spec_requested.channels);
+        fprintf(stderr, "%s:     - samples per frame: %d\n",                   __func__, capture_spec_obtained.samples);
+    }
+
+    m_sample_rate = capture_spec_obtained.freq;
+
+    m_audio.resize((m_sample_rate*m_len_ms)/1000);
+
+    return true;
+}
+
+bool audio_async::resume() {
+    if (!m_dev_id_in) {
+        fprintf(stderr, "%s: no audio device to resume!\n", __func__);
+        return false;
+    }
+
+    if (m_running) {
+        fprintf(stderr, "%s: already running!\n", __func__);
+        return false;
+    }
+
+    SDL_PauseAudioDevice(m_dev_id_in, 0);
+
+    m_running = true;
+
+    return true;
+}
+
+bool audio_async::pause() {
+    if (!m_dev_id_in) {
+        fprintf(stderr, "%s: no audio device to pause!\n", __func__);
+        return false;
+    }
+
+    if (!m_running) {
+        fprintf(stderr, "%s: already paused!\n", __func__);
+        return false;
+    }
+
+    SDL_PauseAudioDevice(m_dev_id_in, 1);
+
+    m_running = false;
+
+    return true;
+}
+
+bool audio_async::clear() {
+    if (!m_dev_id_in) {
+        fprintf(stderr, "%s: no audio device to clear!\n", __func__);
+        return false;
+    }
+
+    if (!m_running) {
+        fprintf(stderr, "%s: not running!\n", __func__);
+        return false;
+    }
+
+    {
+        std::lock_guard<std::mutex> lock(m_mutex);
+
+        m_audio_pos = 0;
+        m_audio_len = 0;
+    }
+
+    return true;
+}
+
+// callback to be called by SDL
+void audio_async::callback(uint8_t * stream, int len) {
+    if (!m_running) {
+        return;
+    }
+
+    size_t n_samples = len / sizeof(float);
+
+    if (n_samples > m_audio.size()) {
+        n_samples = m_audio.size();
+
+        stream += (len - (n_samples * sizeof(float)));
+    }
+
+    //fprintf(stderr, "%s: %zu samples, pos %zu, len %zu\n", __func__, n_samples, m_audio_pos, m_audio_len);
+
+    {
+        std::lock_guard<std::mutex> lock(m_mutex);
+
+        if (m_audio_pos + n_samples > m_audio.size()) {
+            const size_t n0 = m_audio.size() - m_audio_pos;
+
+            memcpy(&m_audio[m_audio_pos], stream, n0 * sizeof(float));
+            memcpy(&m_audio[0], stream + n0 * sizeof(float), (n_samples - n0) * sizeof(float));
+        } else {
+            memcpy(&m_audio[m_audio_pos], stream, n_samples * sizeof(float));
+        }
+        m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
+        m_audio_len = std::min(m_audio_len + n_samples, m_audio.size());
+    }
+}
+
+void audio_async::get(int ms, std::vector<float> & result) {
+    if (!m_dev_id_in) {
+        fprintf(stderr, "%s: no audio device to get audio from!\n", __func__);
+        return;
+    }
+
+    if (!m_running) {
+        fprintf(stderr, "%s: not running!\n", __func__);
+        return;
+    }
+
+    result.clear();
+
+    {
+        std::lock_guard<std::mutex> lock(m_mutex);
+
+        if (ms <= 0) {
+            ms = m_len_ms;
+        }
+
+        size_t n_samples = (m_sample_rate * ms) / 1000;
+        if (n_samples > m_audio_len) {
+            n_samples = m_audio_len;
+        }
+
+        result.resize(n_samples);
+
+        int s0 = m_audio_pos - n_samples;
+        if (s0 < 0) {
+            s0 += m_audio.size();
+        }
+
+        if (s0 + n_samples > m_audio.size()) {
+            const size_t n0 = m_audio.size() - s0;
+
+            memcpy(result.data(), &m_audio[s0], n0 * sizeof(float));
+            memcpy(&result[n0], &m_audio[0], (n_samples - n0) * sizeof(float));
+        } else {
+            memcpy(result.data(), &m_audio[s0], n_samples * sizeof(float));
+        }
+    }
+}
+
+bool sdl_poll_events() {
+    SDL_Event event;
+    while (SDL_PollEvent(&event)) {
+        switch (event.type) {
+            case SDL_QUIT:
+                {
+                    return false;
+                }
+            default:
+                break;
+        }
+    }
+
+    return true;
+}

data/ext/sources/examples/common-sdl.h

@@ -0,0 +1,49 @@
+#pragma once
+
+#include <SDL.h>
+#include <SDL_audio.h>
+
+#include <atomic>
+#include <cstdint>
+#include <vector>
+#include <mutex>
+
+//
+// SDL Audio capture
+//
+
+class audio_async {
+public:
+    audio_async(int len_ms);
+    ~audio_async();
+
+    bool init(int capture_id, int sample_rate);
+
+    // start capturing audio via the provided SDL callback
+    // keep last len_ms seconds of audio in a circular buffer
+    bool resume();
+    bool pause();
+    bool clear();
+
+    // callback to be called by SDL
+    void callback(uint8_t * stream, int len);
+
+    // get audio data from the circular buffer
+    void get(int ms, std::vector<float> & audio);
+
+private:
+    SDL_AudioDeviceID m_dev_id_in = 0;
+
+    int m_len_ms = 0;
+    int m_sample_rate = 0;
+
+    std::atomic_bool m_running;
+    std::mutex       m_mutex;
+
+    std::vector<float> m_audio;
+    size_t             m_audio_pos = 0;
+    size_t             m_audio_len = 0;
+};
+
+// Return false if need to quit
+bool sdl_poll_events();

data/ext/sources/examples/common-whisper.cpp

@@ -0,0 +1,168 @@
+#define _USE_MATH_DEFINES // for M_PI
+
+#include "common-whisper.h"
+
+#include "common.h"
+
+#include "whisper.h"
+
+// third-party utilities
+// use your favorite implementations
+#define STB_VORBIS_HEADER_ONLY
+#include "stb_vorbis.c"    /* Enables Vorbis decoding. */
+
+#ifdef _WIN32
+#ifndef NOMINMAX
+    #define NOMINMAX
+#endif
+#endif
+
+#define MA_NO_DEVICE_IO
+#define MA_NO_THREADING
+#define MA_NO_ENCODING
+#define MA_NO_GENERATION
+#define MA_NO_RESOURCE_MANAGER
+#define MA_NO_NODE_GRAPH
+#define MINIAUDIO_IMPLEMENTATION
+#include "miniaudio.h"
+
+#ifdef _WIN32
+#include <fcntl.h>
+#include <io.h>
+#endif
+
+#include <cstring>
+#include <fstream>
+
+#ifdef WHISPER_FFMPEG
+// as implemented in ffmpeg_trancode.cpp only embedded in common lib if whisper built with ffmpeg support
+extern bool ffmpeg_decode_audio(const std::string & ifname, std::vector<uint8_t> & wav_data);
+#endif
+
+bool read_audio_data(const std::string & fname, std::vector<float>& pcmf32, std::vector<std::vector<float>>& pcmf32s, bool stereo) {
+    std::vector<uint8_t> audio_data; // used for pipe input from stdin or ffmpeg decoding output
+
+    ma_result result;
+    ma_decoder_config decoder_config;
+    ma_decoder decoder;
+
+    decoder_config = ma_decoder_config_init(ma_format_f32, stereo ? 2 : 1, WHISPER_SAMPLE_RATE);
+
+    if (fname == "-") {
+		#ifdef _WIN32
+		_setmode(_fileno(stdin), _O_BINARY);
+		#endif
+
+		uint8_t buf[1024];
+		while (true)
+		{
+			const size_t n = fread(buf, 1, sizeof(buf), stdin);
+			if (n == 0) {
+				break;
+			}
+			audio_data.insert(audio_data.end(), buf, buf + n);
+		}
+
+		if ((result = ma_decoder_init_memory(audio_data.data(), audio_data.size(), &decoder_config, &decoder)) != MA_SUCCESS) {
+
+			fprintf(stderr, "Error: failed to open audio data from stdin (%s)\n", ma_result_description(result));
+
+			return false;
+		}
+
+		fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, audio_data.size());
+    }
+    else if (((result = ma_decoder_init_file(fname.c_str(), &decoder_config, &decoder)) != MA_SUCCESS)) {
+#if defined(WHISPER_FFMPEG)
+		if (ffmpeg_decode_audio(fname, audio_data) != 0) {
+			fprintf(stderr, "error: failed to ffmpeg decode '%s'\n", fname.c_str());
+
+			return false;
+		}
+
+		if ((result = ma_decoder_init_memory(audio_data.data(), audio_data.size(), &decoder_config, &decoder)) != MA_SUCCESS) {
+			fprintf(stderr, "error: failed to read audio data as wav (%s)\n", ma_result_description(result));
+
+			return false;
+		}
+#else
+		if ((result = ma_decoder_init_memory(fname.c_str(), fname.size(), &decoder_config, &decoder)) != MA_SUCCESS) {
+			fprintf(stderr, "error: failed to read audio data as wav (%s)\n", ma_result_description(result));
+
+			return false;
+		}
+#endif
+    }
+
+    ma_uint64 frame_count;
+    ma_uint64 frames_read;
+
+    if ((result = ma_decoder_get_length_in_pcm_frames(&decoder, &frame_count)) != MA_SUCCESS) {
+		fprintf(stderr, "error: failed to retrieve the length of the audio data (%s)\n", ma_result_description(result));
+
+		return false;
+    }
+
+    pcmf32.resize(stereo ? frame_count*2 : frame_count);
+
+    if ((result = ma_decoder_read_pcm_frames(&decoder, pcmf32.data(), frame_count, &frames_read)) != MA_SUCCESS) {
+		fprintf(stderr, "error: failed to read the frames of the audio data (%s)\n", ma_result_description(result));
+
+		return false;
+    }
+
+    if (stereo) {
+		pcmf32s.resize(2);
+		pcmf32s[0].resize(frame_count);
+		pcmf32s[1].resize(frame_count);
+		for (uint64_t i = 0; i < frame_count; i++) {
+			pcmf32s[0][i] = pcmf32[2*i];
+			pcmf32s[1][i] = pcmf32[2*i + 1];
+		}
+    }
+
+    ma_decoder_uninit(&decoder);
+
+    return true;
+}
+
+//  500 -> 00:05.000
+// 6000 -> 01:00.000
+std::string to_timestamp(int64_t t, bool comma) {
+    int64_t msec = t * 10;
+    int64_t hr = msec / (1000 * 60 * 60);
+    msec = msec - hr * (1000 * 60 * 60);
+    int64_t min = msec / (1000 * 60);
+    msec = msec - min * (1000 * 60);
+    int64_t sec = msec / 1000;
+    msec = msec - sec * 1000;
+
+    char buf[32];
+    snprintf(buf, sizeof(buf), "%02d:%02d:%02d%s%03d", (int) hr, (int) min, (int) sec, comma ? "," : ".", (int) msec);
+
+    return std::string(buf);
+}
+
+int timestamp_to_sample(int64_t t, int n_samples, int whisper_sample_rate) {
+    return std::max(0, std::min((int) n_samples - 1, (int) ((t*whisper_sample_rate)/100)));
+}
+
+bool speak_with_file(const std::string & command, const std::string & text, const std::string & path, int voice_id) {
+    std::ofstream speak_file(path.c_str());
+    if (speak_file.fail()) {
+        fprintf(stderr, "%s: failed to open speak_file\n", __func__);
+        return false;
+    } else {
+        speak_file.write(text.c_str(), text.size());
+        speak_file.close();
+        int ret = system((command + " " + std::to_string(voice_id) + " " + path).c_str());
+        if (ret != 0) {
+            fprintf(stderr, "%s: failed to speak\n", __func__);
+            return false;
+        }
+    }
+    return true;
+}
+
+#undef STB_VORBIS_HEADER_ONLY
+#include "stb_vorbis.c"

data/ext/sources/examples/common-whisper.h

@@ -0,0 +1,24 @@
+#pragma once
+
+#include <string>
+#include <vector>
+#include <cstdint>
+
+// Read WAV audio file and store the PCM data into pcmf32
+// fname can be a buffer of WAV data instead of a filename
+// The sample rate of the audio must be equal to COMMON_SAMPLE_RATE
+// If stereo flag is set and the audio has 2 channels, the pcmf32s will contain 2 channel PCM
+bool read_audio_data(
+        const std::string & fname,
+        std::vector<float> & pcmf32,
+        std::vector<std::vector<float>> & pcmf32s,
+        bool stereo);
+
+// convert timestamp to string, 6000 -> 01:00.000
+std::string to_timestamp(int64_t t, bool comma = false);
+
+// given a timestamp get the sample
+int timestamp_to_sample(int64_t t, int n_samples, int whisper_sample_rate);
+
+// write text to file, and call system("command voice_id file")
+bool speak_with_file(const std::string & command, const std::string & text, const std::string & path, int voice_id);