Spaces:
Running
Running
| import streamlit as st | |
| import soundfile as sf | |
| import librosa | |
| import numpy as np | |
| import time | |
| from transformers import pipeline | |
| from io import BytesIO | |
| import tempfile | |
| from streamlit_webrtc import webrtc_streamer, WebRtcMode, RTCConfiguration | |
| import av | |
| import queue | |
| # Define the models | |
| MODELS = { | |
| "Whisper (English)": "openai/whisper-small.en", | |
| "Whisper (Multilingual)": "openai/whisper-small", | |
| "Facebook Wav2Vec2": "facebook/wav2vec2-large-960h", | |
| "Google Wav2Vec2": "google/wav2vec2-large-xlsr-53", | |
| "Whisper (Thai)": "openai/whisper-large" | |
| } | |
| # App UI | |
| st.title("Audio to Text Conversion") | |
| st.subheader("Select language and model") | |
| # Language selection | |
| language = st.selectbox("Choose Language", options=["English", "Thai"]) | |
| # Model selection | |
| model_choice = st.selectbox("Choose a Model", options=list(MODELS.keys())) | |
| # Audio input options | |
| st.subheader("Record or Upload your audio") | |
| audio_option = st.radio("Choose an option:", ('Record Audio', 'Upload Audio')) | |
| audio_data = None | |
| # Queue to store recorded audio frames | |
| audio_queue = queue.Queue() | |
| # WebRTC Audio Recorder | |
| def audio_frame_callback(frame: av.AudioFrame): | |
| audio = frame.to_ndarray() | |
| audio_queue.put(audio) | |
| return frame | |
| # Option 1: Record audio via browser using WebRTC | |
| if audio_option == 'Record Audio': | |
| st.write("Click the button to start/stop recording.") | |
| # Change STUN server to a different one to avoid potential issues | |
| rtc_configuration = RTCConfiguration( | |
| { | |
| "iceServers": [ | |
| {"urls": ["stun:stun1.l.google.com:19302"]}, | |
| {"urls": ["stun:stun2.l.google.com:19302"]} | |
| ] | |
| } | |
| ) | |
| # Start WebRTC recording | |
| webrtc_ctx = webrtc_streamer( | |
| key="audio-stream", | |
| mode=WebRtcMode.SENDONLY, | |
| rtc_configuration=rtc_configuration, | |
| media_stream_constraints={"audio": True, "video": False}, | |
| audio_frame_callback=audio_frame_callback, | |
| ) | |
| # Ensure we are recording | |
| if webrtc_ctx.state.playing: | |
| st.write("Recording...") | |
| # Convert recorded audio frames to a numpy array for processing | |
| recorded_audio = [] | |
| while not audio_queue.empty(): | |
| recorded_audio.append(audio_queue.get()) | |
| if recorded_audio: | |
| audio_data = np.concatenate(recorded_audio, axis=0) | |
| sr = 16000 # Assuming a standard sample rate for WebRTC | |
| # Compute audio properties | |
| audio_size = len(audio_data) * 2 # in bytes (16-bit PCM) | |
| duration = len(audio_data) / sr | |
| # Display audio properties | |
| st.write(f"Audio Size: {audio_size} bytes") | |
| st.write(f"Frame Rate: {sr} Hz") | |
| st.write(f"Duration: {duration:.2f} seconds") | |
| # Perform conversion using the selected model | |
| st.subheader("Converting audio to text...") | |
| start_time = time.time() | |
| # Load the model from HuggingFace | |
| model = pipeline("automatic-speech-recognition", model=MODELS[model_choice]) | |
| # Perform the conversion | |
| audio_bytes = BytesIO(sf.write("temp.wav", audio_data, sr)) | |
| result = model(audio_bytes) | |
| end_time = time.time() | |
| # Display results | |
| st.write("Transcription:", result['text']) | |
| st.write(f"Conversion took {end_time - start_time:.2f} seconds") | |
| # Option 2: Upload audio | |
| elif audio_option == 'Upload Audio': | |
| audio_file = st.file_uploader("Upload audio file (WAV format)", type=['wav']) | |
| if audio_file: | |
| # Load the audio file | |
| with tempfile.NamedTemporaryFile(delete=False) as tmp_file: | |
| tmp_file.write(audio_file.read()) | |
| tmp_file_path = tmp_file.name | |
| audio_data, sr = librosa.load(tmp_file_path, sr=None) | |
| # Compute audio properties | |
| audio_size = len(audio_data) * 2 # in bytes (16-bit PCM) | |
| frame_rate = sr | |
| duration = librosa.get_duration(y=audio_data, sr=sr) | |
| # Display audio properties | |
| st.write(f"Audio Size: {audio_size} bytes") | |
| st.write(f"Frame Rate: {frame_rate} Hz") | |
| st.write(f"Duration: {duration:.2f} seconds") | |
| # Perform conversion using the selected model | |
| st.subheader("Converting audio to text...") | |
| start_time = time.time() | |
| # Load the model from HuggingFace | |
| model = pipeline("automatic-speech-recognition", model=MODELS[model_choice]) | |
| # Perform the conversion | |
| audio_bytes = BytesIO(sf.write(tmp_file_path, audio_data, sr)) | |
| result = model(tmp_file_path) | |
| end_time = time.time() | |
| # Display results | |
| st.write("Transcription:", result['text']) | |
| st.write(f"Conversion took {end_time - start_time:.2f} seconds") | |
| else: | |
| st.write("Please select an audio input option.") |