release : v1.2.1

Just in case

.gitignore

@@ -10,6 +10,7 @@ build-em/
 build-debug/
 build-release/
 build-static/
+build-no-accel/
 build-sanitize-addr/
 build-sanitize-thread/
 

CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required (VERSION 3.0)
 
-project(whisper.cpp VERSION 1.2.0)
+project(whisper.cpp VERSION 1.2.1)
 
 # Add path to modules
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")

Makefile

@@ -141,6 +141,8 @@ ifdef WHISPER_GPROF
 	CXXFLAGS += -pg
 endif
 ifneq ($(filter aarch64%,$(UNAME_M)),)
+	CFLAGS += -mcpu=native
+	CXXFLAGS += -mcpu=native
 endif
 ifneq ($(filter armv6%,$(UNAME_M)),)
 	# Raspberry Pi 1, 2, 3

README.md

@@ -4,7 +4,7 @@
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 [![npm](https://img.shields.io/npm/v/whisper.cpp.svg)](https://www.npmjs.com/package/whisper.cpp/)
 
-Stable: [v1.2.0](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.2.0) / [Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)
+Stable: [v1.2.1](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.2.1) / [Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)
 
 High-performance inference of [OpenAI's Whisper](https://github.com/openai/whisper) automatic speech recognition (ASR) model:
 
@@ -469,7 +469,9 @@ in [models](models).
 - [X] .NET: | [#422](https://github.com/ggerganov/whisper.cpp/discussions/422)
   - [sandrohanea/whisper.net](https://github.com/sandrohanea/whisper.net)
   - [NickDarvey/whisper](https://github.com/NickDarvey/whisper)
-- [ ] Python: soon | [WIP](https://github.com/ggerganov/whisper.cpp/issues/9)
+- [X] Python: | [#9](https://github.com/ggerganov/whisper.cpp/issues/9)
+  - [stlukey/whispercpp.py](https://github.com/stlukey/whispercpp.py) (Cython)
+  - [aarnphm/whispercpp](https://github.com/aarnphm/whispercpp) (Pybind11)
 
 ## Examples
 

bindings/javascript/package.json

@@ -1,6 +1,6 @@
 {
   "name": "whisper.cpp",
-  "version": "1.2.0",
+  "version": "1.2.1",
   "description": "Whisper speech recognition",
   "main": "whisper.js",
   "scripts": {

examples/addon.node/addon.cpp

@@ -72,7 +72,7 @@ int timestamp_to_sample(int64_t t, int n_samples) {
     return std::max(0, std::min((int) n_samples - 1, (int) ((t*WHISPER_SAMPLE_RATE)/100)));
 }
 
-void whisper_print_segment(struct whisper_context * ctx, int n_new, void * user_data) {
+void whisper_print_segment_callback(struct whisper_context * ctx, 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;
 
@@ -250,7 +250,7 @@ int run(whisper_params &params, std::vector<std::vector<std::string>> &result) {
 
             // this callback is called on each new segment
             if (!wparams.print_realtime) {
-                wparams.new_segment_callback           = whisper_print_segment;
+                wparams.new_segment_callback           = whisper_print_segment_callback;
                 wparams.new_segment_callback_user_data = &user_data;
             }
 

examples/command/command.cpp

@@ -109,73 +109,6 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "\n");
 }
 
-struct whisper_logits_filter_user_data {
-    std::vector<std::string>                * allowed_commands;
-    std::vector<std::vector<whisper_token>> * allowed_tokens;
-};
-
-void whisper_logits_filter(
-            struct whisper_context * ctx,
-          const whisper_token_data * tokens,
-                               int   n_tokens,
-                             float * logits,
-                              void * user_data){
-    const auto & allowed_tokens = *((whisper_logits_filter_user_data *) user_data)->allowed_tokens;
-
-    printf("n_tokens = %d\n", n_tokens);
-    for (int i = 0; i < n_tokens; i++) {
-        printf(" - '%s' (%.2f)\n", whisper_token_to_str(ctx, tokens[i].id), logits[i]);
-    }
-
-    if (n_tokens == 0) {
-        return;
-    }
-
-    std::vector<std::pair<whisper_token, float>> pool;
-    for (int i = 0; i < (int) allowed_tokens.size(); i++) {
-        const int n = (int) allowed_tokens[i].size();
-        if (n_tokens > n) {
-            continue;
-        }
-
-        const whisper_token id = allowed_tokens[i][n_tokens - 1];
-        pool.push_back({ id, logits[id] });
-    }
-
-    if (pool.empty()) {
-        return;
-    }
-
-    printf("applying logits filter, pool size = %d\n", (int) pool.size());
-
-    const int ibeg = whisper_token_beg(ctx);
-
-    double sum_all = 0.0;
-    for (int i = 0; i < ibeg; ++i) {
-        if (logits[i] == -INFINITY) {
-            continue;
-        }
-        sum_all += logits[i];
-    }
-
-    double sum_pool = 0.0;
-    for (int i = 0; i < (int) pool.size(); ++i) {
-        sum_pool += pool[i].second;
-    }
-
-    printf("sum_all = %.2f, sum_pool = %.2f\n", sum_all, sum_pool);
-
-    for (int i = 0; i < ibeg; ++i) {
-        logits[i] = -INFINITY;
-    }
-
-    for (int i = 0; i < (int) pool.size(); ++i) {
-        //logits[pool[i].first] = pool[i].second / sum_pool * sum_all;
-        logits[pool[i].first] = pool[i].second;
-        printf(" - '%s' (%.2f)\n", whisper_token_to_str(ctx, pool[i].first), logits[pool[i].first]);
-    }
-}
-
 std::string transcribe(whisper_context * ctx, const whisper_params & params, const std::vector<float> & pcmf32, float & prob, int64_t & t_ms) {
     const auto t_start = std::chrono::high_resolution_clock::now();
 
@@ -198,8 +131,6 @@ std::string transcribe(whisper_context * ctx, const whisper_params & params, con
     wparams.audio_ctx        = params.audio_ctx;
     wparams.speed_up         = params.speed_up;
 
-    wparams.temperature_inc  = -1.0f;
-
     if (whisper_full(ctx, wparams, pcmf32.data(), pcmf32.size()) != 0) {
         return "";
     }
@@ -403,31 +334,22 @@ int process_command_list(struct whisper_context * ctx, audio_async &audio, const
             wparams.translate        = params.translate;
             wparams.no_context       = true;
             wparams.single_segment   = true;
-            //wparams.max_tokens       = 1;
+            wparams.max_tokens       = 1;
             wparams.language         = params.language.c_str();
             wparams.n_threads        = params.n_threads;
 
             wparams.audio_ctx        = params.audio_ctx;
             wparams.speed_up         = params.speed_up;
 
-            wparams.temperature_inc  = -1.0f;
-
             wparams.prompt_tokens    = k_tokens.data();
             wparams.prompt_n_tokens  = k_tokens.size();
 
-            whisper_logits_filter_user_data user_data = { &allowed_commands, &allowed_tokens };
-
-            wparams.logits_filter_callback           = whisper_logits_filter;
-            wparams.logits_filter_callback_user_data = &user_data;
-
             // run the transformer and a single decoding pass
             if (whisper_full(ctx, wparams, pcmf32_cur.data(), pcmf32_cur.size()) != 0) {
                 fprintf(stderr, "%s: ERROR: whisper_full() failed\n", __func__);
                 break;
             }
 
-            fprintf(stdout, "%s: text - '%s'\n", __func__, whisper_full_get_segment_text(ctx, 0));
-
             // estimate command probability
             // NOTE: not optimal
             {
@@ -514,7 +436,7 @@ int process_command_list(struct whisper_context * ctx, audio_async &audio, const
 
 // always-prompt mode
 // transcribe the voice into text after valid prompt
-int process_always_prompt(struct whisper_context * ctx, audio_async & audio, const whisper_params & params) {
+int always_prompt_transcription(struct whisper_context * ctx, audio_async & audio, const whisper_params & params) {
     bool is_running = true;
     bool ask_prompt = true;
 
@@ -574,7 +496,7 @@ int process_always_prompt(struct whisper_context * ctx, audio_async & audio, con
                 const float sim = similarity(prompt, k_prompt);
 
                 //debug
-                //fprintf(stdout, "command size: %d, sim: %f\n", (int) command.size(), sim);
+                //fprintf(stdout, "command size: %i\n", command_length);
 
                 if ((sim > 0.7f) && (command.size() > 0)) {
                     fprintf(stdout, "%s: Command '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", command.c_str(), "\033[0m", (int) t_ms);
@@ -754,7 +676,7 @@ int main(int argc, char ** argv) {
     if (!params.commands.empty()) {
         ret_val = process_command_list(ctx, audio, params);
     } else if (!params.prompt.empty()) {
-        ret_val = process_always_prompt(ctx, audio, params);
+        ret_val = always_prompt_transcription(ctx, audio, params);
     } else {
         ret_val = process_general_transcription(ctx, audio, params);
     }

examples/common-sdl.h

@@ -1,13 +1,13 @@
 #pragma once
 
+#include <SDL.h>
+#include <SDL_audio.h>
+
 #include <atomic>
 #include <cstdint>
 #include <vector>
 #include <mutex>
 
-#include <SDL.h>
-#include <SDL_audio.h>
-
 //
 // SDL Audio capture
 //

examples/main/main.cpp

@@ -193,7 +193,7 @@ struct whisper_print_user_data {
     const std::vector<std::vector<float>> * pcmf32s;
 };
 
-void whisper_print_segment(struct whisper_context * ctx, int n_new, void * user_data) {
+void whisper_print_segment_callback(struct whisper_context * ctx, 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;
 
@@ -597,7 +597,7 @@ int main(int argc, char ** argv) {
 
             // this callback is called on each new segment
             if (!wparams.print_realtime) {
-                wparams.new_segment_callback           = whisper_print_segment;
+                wparams.new_segment_callback           = whisper_print_segment_callback;
                 wparams.new_segment_callback_user_data = &user_data;
             }
 

whisper.cpp

@@ -592,16 +592,16 @@ struct whisper_context {
 
     mutable std::mt19937 rng; // used for sampling at t > 0.0
 
-    int lang_id;
+    int lang_id = 0; // english by default
 
     // [EXPERIMENTAL] token-level timestamps data
-    int64_t t_beg;
-    int64_t t_last;
+    int64_t t_beg = 0;
+    int64_t t_last = 0;
     whisper_token tid_last;
     std::vector<float> energy; // PCM signal energy
 
     // [EXPERIMENTAL] speed-up techniques
-    int32_t exp_n_audio_ctx; // 0 - use default
+    int32_t exp_n_audio_ctx = 0; // 0 - use default
 
     void use_buf(struct ggml_context * ctx, int i) {
 #if defined(WHISPER_USE_SCRATCH)
@@ -4339,7 +4339,7 @@ int whisper_full_n_segments(struct whisper_context * ctx) {
 }
 
 int whisper_full_lang_id(struct whisper_context * ctx) {
-    return ctx->lang_id; 
+    return ctx->lang_id;
 }
 
 int64_t whisper_full_get_segment_t0(struct whisper_context * ctx, int i_segment) {