fix(cpp): str params should not be released early

mybigday · Dec 9, 2023 · 31e09aa · 31e09aa
1 parent 3a78b10
commit 31e09aa
Show file tree

Hide file tree

Showing 5 changed files with 14 additions and 12 deletions.
diff --git a/android/src/main/jni.cpp b/android/src/main/jni.cpp
@@ -236,12 +236,12 @@ struct whisper_full_params createFullParams(JNIEnv *env, jobject options) {
     jstring prompt = readablemap::getString(env, options, "prompt", nullptr);
     if (prompt != nullptr) {
         params.initial_prompt = env->GetStringUTFChars(prompt, nullptr);
-        env->ReleaseStringUTFChars(prompt, params.initial_prompt);
+        env->DeleteLocalRef(prompt);
     }
     jstring language = readablemap::getString(env, options, "language", nullptr);
     if (language != nullptr) {
         params.language = env->GetStringUTFChars(language, nullptr);
-        env->ReleaseStringUTFChars(language, params.language);
+        env->DeleteLocalRef(language);
     }
     return params;
 }
@@ -330,10 +330,10 @@ Java_com_rnwhisper_WhisperContext_createRealtimeTranscribeJob(
     vad.freq_thold = readablemap::getFloat(env, options, "vadFreqThold", 100.0f);
 
     jstring audio_output_path = readablemap::getString(env, options, "audioOutputPath", nullptr);
-    std::string *audio_output_path_str = nullptr;
+    const char* audio_output_path_str = nullptr;
     if (audio_output_path != nullptr) {
-        audio_output_path_str = new std::string(env->GetStringUTFChars(audio_output_path, nullptr));
-        env->ReleaseStringUTFChars(audio_output_path, audio_output_path_str->c_str());
+        audio_output_path_str = env->GetStringUTFChars(audio_output_path, nullptr);
+        env->DeleteLocalRef(audio_output_path);
     }
     job->set_realtime_params(
         vad,
@@ -357,6 +357,7 @@ Java_com_rnwhisper_WhisperContext_finishRealtimeTranscribeJob(
 
     rnwhisper::job *job = rnwhisper::job_get(job_id);
     if (job->audio_output_path != nullptr) {
+        RNWHISPER_LOG_INFO("job->params.language: %s\n", job->params.language);
         std::vector<int> slice_n_samples_vec;
         jint *slice_n_samples_arr = env->GetIntArrayElements(slice_n_samples, nullptr);
         slice_n_samples_vec = std::vector<int>(slice_n_samples_arr, slice_n_samples_arr + env->GetArrayLength(slice_n_samples));
@@ -365,7 +366,7 @@ Java_com_rnwhisper_WhisperContext_finishRealtimeTranscribeJob(
         // TODO: Append in real time so we don't need to keep all slices & also reduce memory usage
         rnaudioutils::save_wav_file(
             rnaudioutils::concat_short_buffers(job->pcm_slices, slice_n_samples_vec),
-            *job->audio_output_path
+            job->audio_output_path
         );
     }
     rnwhisper::job_remove(job_id);

diff --git a/cpp/rn-audioutils.cpp b/cpp/rn-audioutils.cpp
@@ -61,6 +61,8 @@ void save_wav_file(const std::vector<uint8_t>& raw, const std::string& file) {
     output.write(reinterpret_cast<const char*>(data.data()), data.size());
 
     output.close();
+
+    RNWHISPER_LOG_INFO("Saved audio file: %s\n", file.c_str());
 }
 
 } // namespace rnaudioutils
diff --git a/cpp/rn-whisper.cpp b/cpp/rn-whisper.cpp
@@ -63,7 +63,7 @@ void job::set_realtime_params(
     vad_params params,
     int sec,
     int slice_sec,
-    std::string* output_path
+    const char* output_path
 ) {
     vad = params;
     if (vad.vad_ms < 2000) vad.vad_ms = 2000;

diff --git a/cpp/rn-whisper.h b/cpp/rn-whisper.h
@@ -31,9 +31,9 @@ struct job {
     vad_params vad;
     int audio_sec = 0;
     int audio_slice_sec = 0;
-    std::string* audio_output_path = nullptr;
+    const char* audio_output_path = nullptr;
     std::vector<short *> pcm_slices;
-    void set_realtime_params(vad_params vad, int audio_sec, int audio_slice_sec, std::string* audio_output_path);
+    void set_realtime_params(vad_params vad, int sec, int slice_sec, const char* output_path);
     bool vad_simple(int slice_index, int n_samples, int n);
     void put_pcm_data(short* pcm, int slice_index, int n_samples, int n);
     float* pcm_slice_to_f32(int slice_index, int size);

diff --git a/ios/RNWhisperContext.mm b/ios/RNWhisperContext.mm
@@ -119,7 +119,6 @@ - (void)prepareRealtime:(int)jobId options:(NSDictionary *)options {
     self->recordState.sliceNSamples.push_back(0);
 
     self->recordState.job = rnwhisper::job_new(jobId, [self createParams:options jobId:jobId]);
-    std::string audio_output_path = options[@"audioOutputPath"] != nil ? [options[@"audioOutputPath"] UTF8String] : "";
     self->recordState.job->set_realtime_params(
         {
             .use_vad = options[@"useVad"] != nil ? [options[@"useVad"] boolValue] : false,
@@ -129,7 +128,7 @@ - (void)prepareRealtime:(int)jobId options:(NSDictionary *)options {
         },
         options[@"realtimeAudioSec"] != nil ? [options[@"realtimeAudioSec"] intValue] : 0,
         options[@"realtimeAudioSliceSec"] != nil ? [options[@"realtimeAudioSliceSec"] intValue] : 0,
-        options[@"audioOutputPath"] != nil ? &audio_output_path : nullptr
+        options[@"audioOutputPath"] != nil ? [options[@"audioOutputPath"] UTF8String] : nullptr
     );
     self->recordState.isUseSlices = self->recordState.job->audio_slice_sec < self->recordState.job->audio_sec;
 
@@ -227,7 +226,7 @@ - (void)finishRealtimeTranscribe:(RNWhisperContextRecordState*) state result:(NS
         // TODO: Append in real time so we don't need to keep all slices & also reduce memory usage
         rnaudioutils::save_wav_file(
             rnaudioutils::concat_short_buffers(state->job->pcm_slices, state->sliceNSamples),
-            *state->job->audio_output_path
+            state->job->audio_output_path
         );
     }
     state->transcribeHandler(state->job->job_id, @"end", result);