Update app.py

app.py

@@ -1,66 +1,81 @@
-import streamlit as st
+import torch
 from transformers import pipeline
-import numpy as np
-from io import BytesIO
-import torchaudio
-
-# Load the ASR pipeline for speech-to-text
-whisper_asr = pipeline('automatic-speech-recognition', model='facebook/wav2vec2-base-960h')
-
-# Load the text generation pipeline for text-to-text conversion
-text_generator = pipeline("text2text-generation", model="t5-base", tokenizer="t5-base", framework="pt")
-
-# Streamlit UI
-st.title("Automatic Speech Recognition and Text Generation")
-
-# Dropdown menu to select audio file types
-selected_file_type = st.selectbox("Select audio file type", options=["WAV", "MP3"])
-
-# Map selected file type to corresponding file extensions
-file_extensions = {
-    "WAV": ["wav"],
-    "MP3": ["mp3"]
-}
-
-# File uploader with improved UI design
-uploaded_file = st.file_uploader(f"Upload an audio file ({selected_file_type})", type=file_extensions[selected_file_type])
-
-# ASR function to transcribe audio to text
-def transcribe_audio(audio_file):
-    if audio_file is not None:
-        with st.spinner("Transcribing..."):
-            # Read the content of the uploaded file
-            audio_content = audio_file.read()
-            
-            # Load the audio using torchaudio.load
-            waveform, sample_rate = torchaudio.load(BytesIO(audio_content))
-            
-            # Convert to mono if stereo
-            if waveform.shape[0] > 1:
-                waveform = torch.mean(waveform, dim=0, keepdim=True)
-            
-            # Convert the waveform to numpy array
-            waveform_np = waveform.numpy()
-            
-            # Transcribe the audio
-            audio_text = whisper_asr(waveform_np)
-        st.success("Transcription Complete!")
-        return audio_text[0]['transcription']
-
-# Main app logic
-if uploaded_file is not None:
-    st.audio(uploaded_file, format='audio/mp3', start_time=0)
-    if st.button("Transcribe"):
-        transcribed_text = transcribe_audio(uploaded_file)
-        if transcribed_text:
-            st.subheader("Transcription:")
-            st.write(transcribed_text)
-            
-            # Convert the transcribed audio to text
-            with st.spinner("Converting audio to text..."):
-                generated_text = text_generator(transcribed_text)[0]['generated_text']
-            st.success("Text Conversion Complete!")
-            st.subheader("Generated Text:")
-            st.write(generated_text)
+import gradio as gr
+
+MODEL_NAME = "JackismyShephard/whisper-tiny-finetuned-minds14"
+BATCH_SIZE = 8
+
+device = 0 if torch.cuda.is_available() else "cpu"
+
+pipe = pipeline(
+    task="automatic-speech-recognition",
+    model=MODEL_NAME,
+    chunk_length_s=30,
+    device=device,
+)
+
+
+# Copied from https://github.com/openai/whisper/blob/c09a7ae299c4c34c5839a76380ae407e7d785914/whisper/utils.py#L50
+def format_timestamp(
+    seconds: float, always_include_hours: bool = False, decimal_marker: str = "."
+):
+    if seconds is not None:
+        milliseconds = round(seconds * 1000.0)
+
+        hours = milliseconds // 3_600_000
+        milliseconds -= hours * 3_600_000
+
+        minutes = milliseconds // 60_000
+        milliseconds -= minutes * 60_000
+
+        seconds = milliseconds // 1_000
+        milliseconds -= seconds * 1_000
+
+        hours_marker = f"{hours:02d}:" if always_include_hours or hours > 0 else ""
+        return f"{hours_marker}{minutes:02d}:{seconds:02d}{decimal_marker}{milliseconds:03d}"
+    else:
+        # we have a malformed timestamp so just return it as is
+        return seconds
+
+
+def transcribe(file, return_timestamps):
+    outputs = pipe(
+        file,
+        batch_size=BATCH_SIZE,
+        return_timestamps=return_timestamps,
+    )
+    text = outputs["text"]
+    if return_timestamps:
+        timestamps = outputs["chunks"]
+        timestamps = [
+            f"[{format_timestamp(chunk['timestamp'][0])} -> {format_timestamp(chunk['timestamp'][1])}] {chunk['text']}"
+            for chunk in timestamps
+        ]
+        text = "\n".join(str(feature) for feature in timestamps)
+    return text
+
+
+demo = gr.Interface(
+    fn=transcribe,
+    inputs=[
+        gr.Audio(label="Audio", type="filepath"),
+        gr.Checkbox(label="Return timestamps"),
+    ],
+    outputs=gr.Textbox(show_copy_button=True, label="Text"),
+    title="Automatic Speech Recognition",
+    description=(
+        "Transcribe or translate long-form audio file or microphone inputs with the click of a button! Demo uses the"
+        f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe or translate audio"
+        " of arbitrary length."
+    ),
+    examples=[
+        ["examples/example.flac", False],
+        ["examples/example.flac", True],
+    ],
+    cache_examples=True,
+    allow_flagging="never",
+)
+
+demo.launch()