Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
fix(runtime): free custom AVIO buffers on all exits
wangxiuwenc3c66061ee98
diff --git a/runtime/onnxruntime/src/audio.cpp b/runtime/onnxruntime/src/audio.cppindex ba6045f8a..93d6dcd86 100644--- a/runtime/onnxruntime/src/audio.cpp+++ b/runtime/onnxruntime/src/audio.cpp@@ -33,6 +33,18 @@ extern "C" {using namespace std;namespace funasr {+#if !defined(__APPLE__) && defined(ENABLE_FFMPEG)+namespace {+void FreeAvioContext(AVIOContext** avio_ctx) {+ if (avio_ctx == nullptr || *avio_ctx == nullptr) {+ return;+ }+ av_freep(&(*avio_ctx)->buffer);+ avio_context_free(avio_ctx);+}+} // namespace+#endif+// see http://soundfile.sapp.org/doc/WaveFormat/// Note: We assume little endian herestruct WaveHeader {@@ -472,17 +484,17 @@ bool Audio::FfmpegLoad(const char* buf, int n_file_len){formatContext->pb = avio_ctx;if (avformat_open_input(&formatContext, "", nullptr, nullptr) != 0) {LOG(ERROR) << "Error: Could not open input file.";- avio_context_free(&avio_ctx);avformat_close_input(&formatContext);avformat_free_context(formatContext);+ FreeAvioContext(&avio_ctx);return false;}if (avformat_find_stream_info(formatContext, nullptr) < 0) {LOG(ERROR) << "Error: Could not find stream information.";- avio_context_free(&avio_ctx);avformat_close_input(&formatContext);avformat_free_context(formatContext);+ FreeAvioContext(&avio_ctx);return false;}const AVCodec* codec = nullptr;@@ -494,24 +506,24 @@ bool Audio::FfmpegLoad(const char* buf, int n_file_len){AVCodecContext* codecContext = avcodec_alloc_context3(codec);if (!codecContext) {LOG(ERROR) << "Failed to allocate codec context";- avio_context_free(&avio_ctx);avformat_close_input(&formatContext);avformat_free_context(formatContext);+ FreeAvioContext(&avio_ctx);return false;}if (avcodec_parameters_to_context(codecContext, codecParameters) != 0) {LOG(ERROR) << "Error: Could not copy codec parameters to codec context.";- avio_context_free(&avio_ctx);avformat_close_input(&formatContext);avformat_free_context(formatContext);+ FreeAvioContext(&avio_ctx);avcodec_free_context(&codecContext);return false;}if (avcodec_open2(codecContext, codec, nullptr) < 0) {LOG(ERROR) << "Error: Could not open audio decoder.";- avio_context_free(&avio_ctx);avformat_close_input(&formatContext);avformat_free_context(formatContext);+ FreeAvioContext(&avio_ctx);avcodec_free_context(&codecContext);return false;}@@ -528,17 +540,17 @@ bool Audio::FfmpegLoad(const char* buf, int n_file_len){);if (swr_ctx == nullptr) {LOG(ERROR) << "Could not initialize resampler";- avio_context_free(&avio_ctx);avformat_close_input(&formatContext);avformat_free_context(formatContext);+ FreeAvioContext(&avio_ctx);avcodec_free_context(&codecContext);return false;}if (swr_init(swr_ctx) != 0) {LOG(ERROR) << "Could not initialize resampler";- avio_context_free(&avio_ctx);avformat_close_input(&formatContext);avformat_free_context(formatContext);+ FreeAvioContext(&avio_ctx);avcodec_free_context(&codecContext);swr_free(&swr_ctx);return false;@@ -582,11 +594,9 @@ bool Audio::FfmpegLoad(const char* buf, int n_file_len){av_packet_unref(packet);}- //avio_context_free(&avio_ctx);- av_freep(&avio_ctx ->buffer);- av_freep(&avio_ctx);avformat_close_input(&formatContext);avformat_free_context(formatContext);+ FreeAvioContext(&avio_ctx);avcodec_free_context(&codecContext);swr_free(&swr_ctx);av_packet_free(&packet);