whispercpp 1.3.2 → 1.3.4

data/ext/sources/examples/server/server.cpp

@@ -5,6 +5,7 @@
 #include "httplib.h"
 #include "json.hpp"
 
+#include <cfloat>
 #include <chrono>
 #include <cmath>
 #include <cstdio>
@@ -13,10 +14,23 @@
 #include <string>
 #include <thread>
 #include <vector>
+#include <memory>
+#include <csignal>
+#include <atomic>
+#include <functional>
+#include <cstdlib>
+#if defined (_WIN32)
+#include <windows.h>
+#endif
 
 using namespace httplib;
 using json = nlohmann::ordered_json;
 
+enum server_state {
+    SERVER_STATE_LOADING_MODEL,  // Server is starting up, model not fully loaded yet
+    SERVER_STATE_READY,          // Server is ready and model is loaded
+};
+
 namespace {
 
 // output formats
@@ -26,6 +40,20 @@ const std::string srt_format    = "srt";
 const std::string vjson_format  = "verbose_json";
 const std::string vtt_format    = "vtt";
 
+std::function<void(int)> shutdown_handler;
+std::atomic_flag is_terminating = ATOMIC_FLAG_INIT;
+
+inline void signal_handler(int signal) {
+    if (is_terminating.test_and_set()) {
+        // in case it hangs, we can force terminate the server by hitting Ctrl+C twice
+        // this is for better developer experience, we can remove when the server is stable enough
+        fprintf(stderr, "Received second interrupt, terminating immediately.\n");
+        exit(1);
+    }
+
+    shutdown_handler(signal);
+}
+
 struct server_params
 {
     std::string hostname = "127.0.0.1";
@@ -73,9 +101,10 @@ struct whisper_params {
     bool print_progress  = false;
     bool no_timestamps   = false;
     bool use_gpu         = true;
-    bool flash_attn      = false;
+    bool flash_attn      = true;
     bool suppress_nst    = false;
     bool no_context      = false;
+    bool no_language_probabilities = false;
 
     std::string language        = "en";
     std::string prompt          = "";
@@ -90,6 +119,16 @@ struct whisper_params {
     std::string openvino_encode_device = "CPU";
 
     std::string dtw = "";
+
+    // Voice Activity Detection (VAD) parameters
+    bool        vad           = false;
+    std::string vad_model     = "";
+    float       vad_threshold = 0.5f;
+    int         vad_min_speech_duration_ms = 250;
+    int         vad_min_silence_duration_ms = 100;
+    float       vad_max_speech_duration_s = FLT_MAX;
+    int         vad_speech_pad_ms = 30;
+    float       vad_samples_overlap = 0.1f;
 };
 
 void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & params, const server_params& sparams) {
@@ -139,7 +178,21 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -nth N,    --no-speech-thold N [%-7.2f] no speech threshold\n",   params.no_speech_thold);
     fprintf(stderr, "  -nc,       --no-context        [%-7s] do not use previous audio context\n", params.no_context ? "true" : "false");
     fprintf(stderr, "  -ng,       --no-gpu            [%-7s] do not use gpu\n", params.use_gpu ? "false" : "true");
-    fprintf(stderr, "  -fa,       --flash-attn        [%-7s] flash attention\n", params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -fa,       --flash-attn        [%-7s] enable flash attention\n", params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -nfa,      --no-flash-attn     [%-7s] disable flash attention\n", params.flash_attn ? "false" : "true");
+    fprintf(stderr, "  -nlp,      --no-language-probabilities [%-7s] exclude language probabilities from verbose_json output\n", params.no_language_probabilities ? "true" : "false");
+    // Voice Activity Detection (VAD) parameters
+    fprintf(stderr, "\nVoice Activity Detection (VAD) options:\n");
+    fprintf(stderr, "             --vad                           [%-7s] enable Voice Activity Detection (VAD)\n",            params.vad ? "true" : "false");
+    fprintf(stderr, "  -vm FNAME, --vad-model FNAME               [%-7s] VAD model path\n",                                   params.vad_model.c_str());
+    fprintf(stderr, "  -vt N,     --vad-threshold N               [%-7.2f] VAD threshold for speech recognition\n",           params.vad_threshold);
+    fprintf(stderr, "  -vspd N,   --vad-min-speech-duration-ms  N [%-7d] VAD min speech duration (0.0-1.0)\n",                params.vad_min_speech_duration_ms);
+    fprintf(stderr, "  -vsd N,    --vad-min-silence-duration-ms N [%-7d] VAD min silence duration (to split segments)\n",      params.vad_min_silence_duration_ms);
+    fprintf(stderr, "  -vmsd N,   --vad-max-speech-duration-s   N [%-7s] VAD max speech duration (auto-split longer)\n",      params.vad_max_speech_duration_s == FLT_MAX ?
+                                                                                                                                  std::string("FLT_MAX").c_str() :
+                                                                                                                                  std::to_string(params.vad_max_speech_duration_s).c_str());
+    fprintf(stderr, "  -vp N,     --vad-speech-pad-ms           N [%-7d] VAD speech padding (extend segments)\n",             params.vad_speech_pad_ms);
+    fprintf(stderr, "  -vo N,     --vad-samples-overlap         N [%-7.2f] VAD samples overlap (seconds between segments)\n", params.vad_samples_overlap);
     fprintf(stderr, "\n");
 }
 
@@ -184,9 +237,11 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params, serve
         else if (arg == "-dtw"  || arg == "--dtw")             { params.dtw             = argv[++i]; }
         else if (arg == "-ng"   || arg == "--no-gpu")          { params.use_gpu         = false; }
         else if (arg == "-fa"   || arg == "--flash-attn")      { params.flash_attn      = true; }
+        else if (arg == "-nfa"  || arg == "--no-flash-attn")   { params.flash_attn      = false; }
         else if (arg == "-sns"  || arg == "--suppress-nst")    { params.suppress_nst    = true; }
         else if (arg == "-nth"  || arg == "--no-speech-thold") { params.no_speech_thold = std::stof(argv[++i]); }
         else if (arg == "-nc"   || arg == "--no-context")      { params.no_context      = true; }
+        else if (arg == "-nlp"  || arg == "--no-language-probabilities") { params.no_language_probabilities = true; }
 
         // server params
         else if (                  arg == "--port")            { sparams.port        = std::stoi(argv[++i]); }
@@ -195,6 +250,16 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params, serve
         else if (                  arg == "--request-path")    { sparams.request_path = argv[++i]; }
         else if (                  arg == "--inference-path")  { sparams.inference_path = argv[++i]; }
         else if (                  arg == "--convert")         { sparams.ffmpeg_converter     = true; }
+
+        // Voice Activity Detection (VAD)
+        else if (                  arg == "--vad")                         { params.vad                         = true; }
+        else if (arg == "-vm"   || arg == "--vad-model")                   { params.vad_model                   = argv[++i]; }
+        else if (arg == "-vt"   || arg == "--vad-threshold")               { params.vad_threshold               = std::stof(argv[++i]); }
+        else if (arg == "-vspd" || arg == "--vad-min-speech-duration-ms")  { params.vad_min_speech_duration_ms  = std::stoi(argv[++i]); }
+        else if (arg == "-vsd"  || arg == "--vad-min-silence-duration-ms") { params.vad_min_speech_duration_ms  = std::stoi(argv[++i]); }
+        else if (arg == "-vmsd" || arg == "--vad-max-speech-duration-s")   { params.vad_max_speech_duration_s   = std::stof(argv[++i]); }
+        else if (arg == "-vp"   || arg == "--vad-speech-pad-ms")           { params.vad_speech_pad_ms           = std::stoi(argv[++i]); }
+        else if (arg == "-vo"   || arg == "--vad-samples-overlap")         { params.vad_samples_overlap         = std::stof(argv[++i]); }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params, sparams);
@@ -511,11 +576,45 @@ void get_req_parameters(const Request & req, whisper_params & params)
     {
         params.no_context = parse_str_to_bool(req.get_file_value("no_context").content);
     }
+    if (req.has_file("vad"))
+    {
+        params.vad = parse_str_to_bool(req.get_file_value("vad").content);
+    }
+    if (req.has_file("vad_threshold"))
+    {
+        params.vad_threshold = std::stof(req.get_file_value("vad_threshold").content);
+    }
+    if (req.has_file("vad_min_speech_duration_ms"))
+    {
+        params.vad_min_speech_duration_ms = std::stof(req.get_file_value("vad_min_speech_duration_ms").content);
+    }
+    if (req.has_file("vad_min_silence_duration_ms"))
+    {
+        params.vad_min_silence_duration_ms = std::stof(req.get_file_value("vad_min_silence_duration_ms").content);
+    }
+    if (req.has_file("vad_max_speech_duration_s"))
+    {
+        params.vad_max_speech_duration_s = std::stof(req.get_file_value("vad_max_speech_duration_s").content);
+    }
+    if (req.has_file("vad_speech_pad_ms"))
+    {
+        params.vad_speech_pad_ms = std::stoi(req.get_file_value("vad_speech_pad_ms").content);
+    }
+    if (req.has_file("vad_samples_overlap"))
+    {
+        params.vad_samples_overlap = std::stof(req.get_file_value("vad_samples_overlap").content);
+    }
+    if (req.has_file("no_language_probabilities"))
+    {
+        params.no_language_probabilities = parse_str_to_bool(req.get_file_value("no_language_probabilities").content);
+    }
 }
 
 }  // namespace
 
 int main(int argc, char ** argv) {
+    ggml_backend_load_all();
+
     whisper_params params;
     server_params sparams;
 
@@ -584,13 +683,19 @@ int main(int argc, char ** argv) {
         if (params.dtw == "large.v3") {
             cparams.dtw_aheads_preset = WHISPER_AHEADS_LARGE_V3;
         }
-
+        if (params.dtw == "large.v3.turbo") { 
+            cparams.dtw_aheads_preset = WHISPER_AHEADS_LARGE_V3_TURBO;
+        }
+        
         if (cparams.dtw_aheads_preset == WHISPER_AHEADS_NONE) {
             fprintf(stderr, "error: unknown DTW preset '%s'\n", params.dtw.c_str());
             return 3;
         }
     }
 
+    std::unique_ptr<httplib::Server> svr = std::make_unique<httplib::Server>();
+    std::atomic<server_state> state{SERVER_STATE_LOADING_MODEL};
+
     struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
 
     if (ctx == nullptr) {
@@ -600,9 +705,10 @@ int main(int argc, char ** argv) {
 
     // initialize openvino encoder. this has no effect on whisper.cpp builds that don't have OpenVINO configured
     whisper_ctx_init_openvino_encoder(ctx, nullptr, params.openvino_encode_device.c_str(), nullptr);
+    state.store(SERVER_STATE_READY);
 
-    Server svr;
-    svr.set_default_headers({{"Server", "whisper.cpp"},
+
+    svr->set_default_headers({{"Server", "whisper.cpp"},
                              {"Access-Control-Allow-Origin", "*"},
                              {"Access-Control-Allow-Headers", "content-type, authorization"}});
 
@@ -681,15 +787,15 @@ int main(int argc, char ** argv) {
     whisper_params default_params = params;
 
     // this is only called if no index.html is found in the public --path
-    svr.Get(sparams.request_path + "/", [&default_content](const Request &, Response &res){
+    svr->Get(sparams.request_path + "/", [&](const Request &, Response &res){
         res.set_content(default_content, "text/html");
         return false;
     });
 
-    svr.Options(sparams.request_path + sparams.inference_path, [&](const Request &, Response &){
+    svr->Options(sparams.request_path + sparams.inference_path, [&](const Request &, Response &){
     });
 
-    svr.Post(sparams.request_path + sparams.inference_path, [&](const Request &req, Response &res){
+    svr->Post(sparams.request_path + sparams.inference_path, [&](const Request &req, Response &res){
         // acquire whisper model mutex lock
         std::lock_guard<std::mutex> lock(whisper_mutex);
 
@@ -827,6 +933,16 @@ int main(int argc, char ** argv) {
 
             wparams.suppress_nst     = params.suppress_nst;
 
+            wparams.vad              = params.vad;
+            wparams.vad_model_path   = params.vad_model.c_str();
+
+            wparams.vad_params.threshold               = params.vad_threshold;
+            wparams.vad_params.min_speech_duration_ms  = params.vad_min_speech_duration_ms;
+            wparams.vad_params.min_silence_duration_ms = params.vad_min_silence_duration_ms;
+            wparams.vad_params.max_speech_duration_s   = params.vad_max_speech_duration_s;
+            wparams.vad_params.speech_pad_ms           = params.vad_speech_pad_ms;
+            wparams.vad_params.samples_overlap         = params.vad_samples_overlap;
+
             whisper_print_user_data user_data = { &params, &pcmf32s, 0 };
 
             // this callback is called on each new segment
@@ -917,23 +1033,25 @@ int main(int argc, char ** argv) {
         } else if (params.response_format == vjson_format) {
             /* try to match openai/whisper's Python format */
             std::string results = output_str(ctx, params, pcmf32s); 
-            // Get language probabilities
-            std::vector<float> lang_probs(whisper_lang_max_id() + 1, 0.0f);
-            const auto detected_lang_id = whisper_lang_auto_detect(ctx, 0, params.n_threads, lang_probs.data());
             json jres = json{
                 {"task", params.translate ? "translate" : "transcribe"},
                 {"language", whisper_lang_str_full(whisper_full_lang_id(ctx))},
                 {"duration", float(pcmf32.size())/WHISPER_SAMPLE_RATE},
                 {"text", results},
-                {"segments", json::array()},
-                {"detected_language", whisper_lang_str_full(detected_lang_id)},
-                {"detected_language_probability", lang_probs[detected_lang_id]},
-                {"language_probabilities", json::object()}
+                {"segments", json::array()}
             };
-            // Add all language probabilities
-            for (int i = 0; i <= whisper_lang_max_id(); ++i) {
-                if (lang_probs[i] > 0.001f) { // Only include non-negligible probabilities
-                    jres["language_probabilities"][whisper_lang_str(i)] = lang_probs[i];
+            // Only compute language probabilities if requested (expensive operation)
+            if (!params.no_language_probabilities) {
+                std::vector<float> lang_probs(whisper_lang_max_id() + 1, 0.0f);
+                const auto detected_lang_id = whisper_lang_auto_detect(ctx, 0, params.n_threads, lang_probs.data());
+                jres["detected_language"] = whisper_lang_str_full(detected_lang_id);
+                jres["detected_language_probability"] = lang_probs[detected_lang_id];
+                jres["language_probabilities"] = json::object();
+                // Add all language probabilities
+                for (int i = 0; i <= whisper_lang_max_id(); ++i) {
+                    if (lang_probs[i] > 0.001f) { // Only include non-negligible probabilities
+                        jres["language_probabilities"][whisper_lang_str(i)] = lang_probs[i];
+                    }
                 }
             }
             const int n_segments = whisper_full_n_segments(ctx);
@@ -995,8 +1113,9 @@ int main(int argc, char ** argv) {
         // reset params to their defaults
         params = default_params;
     });
-    svr.Post(sparams.request_path + "/load", [&](const Request &req, Response &res){
+    svr->Post(sparams.request_path + "/load", [&](const Request &req, Response &res){
         std::lock_guard<std::mutex> lock(whisper_mutex);
+        state.store(SERVER_STATE_LOADING_MODEL);
         if (!req.has_file("model"))
         {
             fprintf(stderr, "error: no 'model' field in the request\n");
@@ -1028,18 +1147,25 @@ int main(int argc, char ** argv) {
         // initialize openvino encoder. this has no effect on whisper.cpp builds that don't have OpenVINO configured
         whisper_ctx_init_openvino_encoder(ctx, nullptr, params.openvino_encode_device.c_str(), nullptr);
 
+        state.store(SERVER_STATE_READY);
         const std::string success = "Load was successful!";
         res.set_content(success, "application/text");
 
         // check if the model is in the file system
     });
 
-    svr.Get(sparams.request_path + "/health", [&](const Request &, Response &res){
-        const std::string health_response = "{\"status\":\"ok\"}";
-        res.set_content(health_response, "application/json");
+    svr->Get(sparams.request_path + "/health", [&](const Request &, Response &res){
+        server_state current_state = state.load();
+        if (current_state == SERVER_STATE_READY) {
+            const std::string health_response = "{\"status\":\"ok\"}";
+            res.set_content(health_response, "application/json");
+        } else {
+            res.set_content("{\"status\":\"loading model\"}", "application/json");
+            res.status = 503;
+        }
     });
 
-    svr.set_exception_handler([](const Request &, Response &res, std::exception_ptr ep) {
+    svr->set_exception_handler([](const Request &, Response &res, std::exception_ptr ep) {
         const char fmt[] = "500 Internal Server Error\n%s";
         char buf[BUFSIZ];
         try {
@@ -1053,7 +1179,7 @@ int main(int argc, char ** argv) {
         res.status = 500;
     });
 
-    svr.set_error_handler([](const Request &req, Response &res) {
+    svr->set_error_handler([](const Request &req, Response &res) {
         if (res.status == 400) {
             res.set_content("Invalid request", "text/plain");
         } else if (res.status != 500) {
@@ -1063,10 +1189,10 @@ int main(int argc, char ** argv) {
     });
 
     // set timeouts and change hostname and port
-    svr.set_read_timeout(sparams.read_timeout);
-    svr.set_write_timeout(sparams.write_timeout);
+    svr->set_read_timeout(sparams.read_timeout);
+    svr->set_write_timeout(sparams.write_timeout);
 
-    if (!svr.bind_to_port(sparams.hostname, sparams.port))
+    if (!svr->bind_to_port(sparams.hostname, sparams.port))
     {
         fprintf(stderr, "\ncouldn't bind to server socket: hostname=%s port=%d\n\n",
                 sparams.hostname.c_str(), sparams.port);
@@ -1074,18 +1200,50 @@ int main(int argc, char ** argv) {
     }
 
     // Set the base directory for serving static files
-    svr.set_base_dir(sparams.public_path);
+    svr->set_base_dir(sparams.public_path);
 
     // to make it ctrl+clickable:
     printf("\nwhisper server listening at http://%s:%d\n\n", sparams.hostname.c_str(), sparams.port);
 
-    if (!svr.listen_after_bind())
-    {
-        return 1;
-    }
+    shutdown_handler = [&](int signal) {
+        printf("\nCaught signal %d, shutting down gracefully...\n", signal);
+        if (svr) {
+            svr->stop();
+        }
+    };
+
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
+    struct sigaction sigint_action;
+    sigint_action.sa_handler = signal_handler;
+    sigemptyset (&sigint_action.sa_mask);
+    sigint_action.sa_flags = 0;
+    sigaction(SIGINT, &sigint_action, NULL);
+    sigaction(SIGTERM, &sigint_action, NULL);
+#elif defined (_WIN32)
+    auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL {
+        return (ctrl_type == CTRL_C_EVENT) ? (signal_handler(SIGINT), true) : false;
+    };
+    SetConsoleCtrlHandler(reinterpret_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true);
+#endif
+
+    // clean up function, to be called before exit
+    auto clean_up = [&]() {
+        whisper_print_timings(ctx);
+        whisper_free(ctx);
+    };
+
+    std::thread t([&] {
+        if (!svr->listen_after_bind()) {
+            fprintf(stderr, "error: server listen failed\n");
+        }
+    });
+
+    svr->wait_until_ready();
+
+    t.join();
+
 
-    whisper_print_timings(ctx);
-    whisper_free(ctx);
+    clean_up();
 
     return 0;
 }

data/ext/sources/examples/server.py

@@ -47,7 +47,12 @@ class CustomHTTPRequestHandler(http.server.SimpleHTTPRequestHandler):
 
             elif actual_path == '/':
                 self.path = '/whisper.wasm/index.html'
-            elif actual_path.startswith('/bench.wasm/') or actual_path.startswith('/command.wasm/') or actual_path.startswith('/stream.wasm/'):
+            elif any(actual_path.startswith(prefix) for prefix in (
+                '/bench.wasm/',
+                '/command.wasm/',
+                '/stream.wasm/',
+                '/wchess.wasm/'
+            )):
                 # Keep the path as is, just remove the context root
                 self.path = actual_path
             # For all other paths under the context root

data/ext/sources/examples/stream/stream.cpp

@@ -36,7 +36,7 @@ struct whisper_params {
     bool tinydiarize   = false;
     bool save_audio    = false; // save audio to wav file
     bool use_gpu       = true;
-    bool flash_attn    = false;
+    bool flash_attn    = true;
 
     std::string language  = "en";
     std::string model     = "models/ggml-base.en.bin";
@@ -74,6 +74,7 @@ static bool whisper_params_parse(int argc, char ** argv, whisper_params & params
         else if (arg == "-sa"   || arg == "--save-audio")    { params.save_audio    = true; }
         else if (arg == "-ng"   || arg == "--no-gpu")        { params.use_gpu       = false; }
         else if (arg == "-fa"   || arg == "--flash-attn")    { params.flash_attn    = true; }
+        else if (arg == "-nfa"  || arg == "--no-flash-attn") { params.flash_attn    = false; }
 
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
@@ -111,11 +112,14 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -tdrz,    --tinydiarize   [%-7s] enable tinydiarize (requires a tdrz model)\n",     params.tinydiarize ? "true" : "false");
     fprintf(stderr, "  -sa,      --save-audio    [%-7s] save the recorded audio to a file\n",              params.save_audio ? "true" : "false");
     fprintf(stderr, "  -ng,      --no-gpu        [%-7s] disable GPU inference\n",                          params.use_gpu ? "false" : "true");
-    fprintf(stderr, "  -fa,      --flash-attn    [%-7s] flash attention during inference\n",               params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -fa,      --flash-attn    [%-7s] enable flash attention during inference\n",        params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -nfa,     --no-flash-attn [%-7s] disable flash attention during inference\n",       params.flash_attn ? "false" : "true");
     fprintf(stderr, "\n");
 }
 
 int main(int argc, char ** argv) {
+    ggml_backend_load_all();
+
     whisper_params params;
 
     if (whisper_params_parse(argc, argv, params) == false) {
@@ -161,6 +165,10 @@ int main(int argc, char ** argv) {
     cparams.flash_attn = params.flash_attn;
 
     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 2;
+    }
 
     std::vector<float> pcmf32    (n_samples_30s, 0.0f);
     std::vector<float> pcmf32_old;

data/ext/sources/examples/stream.wasm/emscripten.cpp

@@ -31,10 +31,11 @@ void stream_set_status(const std::string & status) {
     g_status = status;
 }
 
-void stream_main(size_t index) {
+void stream_main(size_t index, const std::string & lang) {
     stream_set_status("loading data ...");
 
     struct whisper_full_params wparams = whisper_full_default_params(whisper_sampling_strategy::WHISPER_SAMPLING_GREEDY);
+    bool is_multilingual = whisper_is_multilingual(g_contexts[index]);
 
     wparams.n_threads        = std::min(N_THREAD, (int) std::thread::hardware_concurrency());
     wparams.offset_ms        = 0;
@@ -52,7 +53,7 @@ void stream_main(size_t index) {
     // disable temperature fallback
     wparams.temperature_inc  = -1.0f;
 
-    wparams.language         = "en";
+    wparams.language         = is_multilingual ? lang.c_str() : "en";
 
     printf("stream: using %d threads\n", wparams.n_threads);
 
@@ -127,9 +128,8 @@ void stream_main(size_t index) {
         g_contexts[index] = nullptr;
     }
 }
-
 EMSCRIPTEN_BINDINGS(stream) {
-    emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
+    emscripten::function("init", emscripten::optional_override([](const std::string & path_model, const std::string & lang) {
         for (size_t i = 0; i < g_contexts.size(); ++i) {
             if (g_contexts[i] == nullptr) {
                 g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
@@ -138,8 +138,8 @@ EMSCRIPTEN_BINDINGS(stream) {
                     if (g_worker.joinable()) {
                         g_worker.join();
                     }
-                    g_worker = std::thread([i]() {
-                        stream_main(i);
+                    g_worker = std::thread([i, lang]() {
+                        stream_main(i, lang);
                     });
 
                     return i + 1;

data/ext/sources/examples/stream.wasm/index-tmpl.html

@@ -42,6 +42,7 @@
                 <a href="../bench.wasm/">bench</a> |
                 <a href="../stream.wasm">stream</a> |
                 <a href="../command.wasm/">command</a> |
+                <a href="../wchess.wasm/">wchess</a> |
 
             <br><br>
 
@@ -55,6 +56,7 @@
                 Whisper model: <span id="model-whisper-status"></span>
                 <button id="fetch-whisper-tiny-en" onclick="loadWhisper('tiny.en')">tiny.en (75 MB)</button>
                 <button id="fetch-whisper-base-en" onclick="loadWhisper('base.en')">base.en (142 MB)</button>
+                <button id="fetch-whisper-base" onclick="loadWhisper('base')">base (142 MB)</button>
                 <br><br>
                 Quantized models:<br><br>
                 <button id="fetch-whisper-tiny-en-q5_1"   onclick="loadWhisper('tiny-en-q5_1')">tiny.en (Q5_1, 31 MB)</button>
@@ -66,6 +68,77 @@
                 -->
             </div>
 
+            <table>
+                <tr>
+                    <td>
+                        Language:
+                        <select id="language" name="language">
+                            <option value="en">English</option>
+                            <option value="ar">Arabic</option>
+                            <option value="hy">Armenian</option>
+                            <option value="az">Azerbaijani</option>
+                            <option value="eu">Basque</option>
+                            <option value="be">Belarusian</option>
+                            <option value="bn">Bengali</option>
+                            <option value="bg">Bulgarian</option>
+                            <option value="ca">Catalan</option>
+                            <option value="zh">Chinese</option>
+                            <option value="hr">Croatian</option>
+                            <option value="cs">Czech</option>
+                            <option value="da">Danish</option>
+                            <option value="nl">Dutch</option>
+                            <option value="en">English</option>
+                            <option value="et">Estonian</option>
+                            <option value="tl">Filipino</option>
+                            <option value="fi">Finnish</option>
+                            <option value="fr">French</option>
+                            <option value="gl">Galician</option>
+                            <option value="ka">Georgian</option>
+                            <option value="de">German</option>
+                            <option value="el">Greek</option>
+                            <option value="gu">Gujarati</option>
+                            <option value="iw">Hebrew</option>
+                            <option value="hi">Hindi</option>
+                            <option value="hu">Hungarian</option>
+                            <option value="is">Icelandic</option>
+                            <option value="id">Indonesian</option>
+                            <option value="ga">Irish</option>
+                            <option value="it">Italian</option>
+                            <option value="ja">Japanese</option>
+                            <option value="kn">Kannada</option>
+                            <option value="ko">Korean</option>
+                            <option value="la">Latin</option>
+                            <option value="lv">Latvian</option>
+                            <option value="lt">Lithuanian</option>
+                            <option value="mk">Macedonian</option>
+                            <option value="ms">Malay</option>
+                            <option value="mt">Maltese</option>
+                            <option value="no">Norwegian</option>
+                            <option value="fa">Persian</option>
+                            <option value="pl">Polish</option>
+                            <option value="pt">Portuguese</option>
+                            <option value="ro">Romanian</option>
+                            <option value="ru">Russian</option>
+                            <option value="sr">Serbian</option>
+                            <option value="sk">Slovak</option>
+                            <option value="sl">Slovenian</option>
+                            <option value="es">Spanish</option>
+                            <option value="sw">Swahili</option>
+                            <option value="sv">Swedish</option>
+                            <option value="ta">Tamil</option>
+                            <option value="te">Telugu</option>
+                            <option value="th">Thai</option>
+                            <option value="tr">Turkish</option>
+                            <option value="uk">Ukrainian</option>
+                            <option value="ur">Urdu</option>
+                            <option value="vi">Vietnamese</option>
+                            <option value="cy">Welsh</option>
+                            <option value="yi">Yiddish</option>
+                        </select>
+                    </td>
+                </tr>
+            </table>
+
             <br>
 
             <div id="input">
@@ -174,16 +247,18 @@
 
             function loadWhisper(model) {
                 let urls = {
-                    'tiny.en': 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en.bin',
-                    'base.en': 'https://whisper.ggerganov.com/ggml-model-whisper-base.en.bin',
+                    'tiny.en': 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en.bin',
+                    'base.en': 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin',
+                    'base'   : 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin',
 
-                    'tiny-en-q5_1':  'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en-q5_1.bin',
-                    'base-en-q5_1':  'https://whisper.ggerganov.com/ggml-model-whisper-base.en-q5_1.bin',
+                    'tiny-en-q5_1':  'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en-q5_1.bin',
+                    'base-en-q5_1':  'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en-q5_1.bin',
                 };
 
                 let sizes = {
                     'tiny.en': 75,
                     'base.en': 142,
+                    'base':     142,
 
                     'tiny-en-q5_1':   31,
                     'base-en-q5_1':   57,
@@ -197,6 +272,7 @@
 
                 document.getElementById('fetch-whisper-tiny-en').style.display = 'none';
                 document.getElementById('fetch-whisper-base-en').style.display = 'none';
+                document.getElementById('fetch-whisper-base').style.display = 'none';
 
                 document.getElementById('fetch-whisper-tiny-en-q5_1').style.display = 'none';
                 document.getElementById('fetch-whisper-base-en-q5_1').style.display = 'none';
@@ -212,6 +288,7 @@
                     var el;
                     el = document.getElementById('fetch-whisper-tiny-en'); if (el) el.style.display = 'inline-block';
                     el = document.getElementById('fetch-whisper-base-en'); if (el) el.style.display = 'inline-block';
+                    el = document.getElementById('fetch-whisper-base'); if (el) el.style.display = 'inline-block';
 
                     el = document.getElementById('fetch-whisper-tiny-en-q5_1'); if (el) el.style.display = 'inline-block';
                     el = document.getElementById('fetch-whisper-base-en-q5_1'); if (el) el.style.display = 'inline-block';
@@ -368,7 +445,7 @@
 
             function onStart() {
                 if (!instance) {
-                    instance = Module.init('whisper.bin');
+                    instance = Module.init('whisper.bin', document.getElementById('language').value);
 
                     if (instance) {
                         printTextarea("js: whisper initialized, instance: " + instance);

data/ext/sources/examples/talk-llama/CMakeLists.txt

@@ -17,6 +17,9 @@ if (WHISPER_SDL2)
         llama-impl.cpp
         llama-io.cpp
         llama-kv-cache.cpp
+        llama-kv-cache-iswa.cpp
+        llama-memory-recurrent.cpp
+        llama-memory-hybrid.cpp
         llama-memory.cpp
         llama-mmap.cpp
         llama-model-loader.cpp