UltraMarkoBR/OpenAI_TTS_Streaming

1import gradio as gr2import numpy as np3import io4import os5from openai import OpenAI6from pydub import AudioSegment7from pydub.playback import play8 9 10# Set an environment variable for key11os.environ['OPENAI_API_KEY'] = os.environ.get('OPENAI_API_KEY')12 13client = OpenAI() # add api_key14 15def stream_and_yield_audio(text, model, voice):16    response = client.audio.speech.create(17        model=model, #"tts-1", for example18        voice=voice , #"alloy", for example19        input=text,20    )21 22    # Convert the binary response content to a byte stream23    byte_stream = io.BytesIO(response.content)24 25    # Read the audio data from the byte stream26    audio = AudioSegment.from_file(byte_stream, format="mp3")27 28    # Export the audio as WAV format29    sample_width = audio.sample_width30    sample_rate = audio.frame_rate31    audio_data = np.array(audio.get_array_of_samples(), dtype=np.int16)32 33    # Yield the audio data34    yield sample_rate, audio_data #audio_data.tobytes(), sample_width35    36 37# demo using older gradio version (3.50.2) 38with gr.Blocks() as demo:39    with gr.Row():40      model = gr.Dropdown(choices=['tts-1','tts-1-hd'], label='Model', value='tts-1')41      voice = gr.Dropdown(choices=['alloy', 'echo', 'fable', 'onyx', 'nova', 'shimmer'], label='Voice Options', value='alloy')42 43    text = gr.Textbox(label="Input text")44    btn = gr.Button("Greet")45    output_audio = gr.Audio(label="Speech Output", streaming=True, autoplay=True)46    47    btn.click(fn=stream_and_yield_audio, inputs=[text,model, voice], outputs=output_audio, api_name="tts-stream")48 49demo.queue().launch()