fix

Author: michalkulakowski

src/audio/audio_utils.cpp

@@ -163,8 +163,8 @@ ov::genai::RawSpeechInput readMp3(const std::string_view& mp3Data) {
     std::vector<float> output(buffer, buffer + outputLength);
     return output;
 }
-#pragma warning(pop)
-void prepareAudioOutput(void** ppData, size_t& pDataSize, uint16_t bitsPerSample, size_t speechSize, ov::Tensor& cpuTensor) {
+
+void prepareAudioOutput(void** ppData, size_t& pDataSize, uint16_t bitsPerSample, size_t speechSize, const float* waveformPtr) {
     enum : unsigned int {
         OUTPUT_PREPARATION,
         TIMER_END
@@ -181,7 +181,6 @@ void prepareAudioOutput(void** ppData, size_t& pDataSize, uint16_t bitsPerSample
     auto waveformSize = speechSize;
     size_t totalSamples = waveformSize * format.channels;
 
-    auto waveformPtr = cpuTensor.data<const float>();
     OPENVINO_ASSERT(drwav_init_memory_write_sequential_pcm_frames(&wav, ppData, &pDataSize, &format, totalSamples, nullptr),
         "Failed to initialize WAV writer");
     drwav_uint64 framesWritten = drwav_write_pcm_frames(&wav, totalSamples, waveformPtr);

src/audio/audio_utils.hpp

@@ -25,4 +25,4 @@ bool isWavBuffer(const std::string buf);
 
 ov::genai::RawSpeechInput readWav(const std::string_view& wavData);
 ov::genai::RawSpeechInput readMp3(const std::string_view& mp3Data);
-void prepareAudioOutput(void** ppData, size_t& pDataSize, uint16_t bitsPerSample, size_t speechSize, ov::Tensor& cpuTensor);
+void prepareAudioOutput(void** ppData, size_t& pDataSize, uint16_t bitsPerSample, size_t speechSize, const float* waveformPtr);

src/audio/text_to_speech/tts_calculator.cc

@@ -108,13 +108,13 @@ class TtsCalculator : public CalculatorBase {
             auto generatedSpeech = pipe->ttsPipeline->generate(inputIt->value.GetString());
             auto bitsPerSample = generatedSpeech.speeches[0].get_element_type().bitwidth();
             auto speechSize = generatedSpeech.speeches[0].get_size();
-            ov::Tensor cpu_tensor(generatedSpeech.speeches[0].get_element_type(), generatedSpeech.speeches[0].get_shape());
+            ov::Tensor cpuTensor(generatedSpeech.speeches[0].get_element_type(), generatedSpeech.speeches[0].get_shape());
             // copy results to release inference request
-            generatedSpeech.speeches[0].copy_to(cpu_tensor);
+            generatedSpeech.speeches[0].copy_to(cpuTensor);
             lock.unlock();
             void* ppData;
             size_t pDataSize;
-            prepareAudioOutput(&ppData, pDataSize, bitsPerSample, speechSize, cpu_tensor);
+            prepareAudioOutput(&ppData, pDataSize, bitsPerSample, speechSize, cpuTensor.data<const float>());
             output = std::make_unique<std::string>(reinterpret_cast<char*>(ppData), pDataSize);
             // drwav_free(ppData, NULL); TODO: is needed?
         } else {