root
commited on
Commit
Β·
a5f1898
1
Parent(s):
0fde622
compare
Browse files- app.py +68 -0
- requirements.txt +3 -0
app.py
ADDED
|
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
from transformers import pipeline
|
| 3 |
+
|
| 4 |
+
# Preload both models
|
| 5 |
+
models = {
|
| 6 |
+
"moulsot_v0.1_2500": pipeline("automatic-speech-recognition", model="01Yassine/moulsot_v0.1_2500"),
|
| 7 |
+
"moulsot_v0.2_1000": pipeline("automatic-speech-recognition", model="01Yassine/moulsot_v0.2_1000")
|
| 8 |
+
}
|
| 9 |
+
|
| 10 |
+
# Adjust generation config for both
|
| 11 |
+
for m in models.values():
|
| 12 |
+
m.model.generation_config.input_ids = m.model.generation_config.forced_decoder_ids
|
| 13 |
+
m.model.generation_config.forced_decoder_ids = None
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
def transcribe(audio, selected_model):
|
| 17 |
+
if audio is None:
|
| 18 |
+
return "Please record or upload an audio file.", "Please record or upload an audio file."
|
| 19 |
+
|
| 20 |
+
pipe = models[selected_model]
|
| 21 |
+
other_model = [k for k in models if k != selected_model][0]
|
| 22 |
+
|
| 23 |
+
# Run inference
|
| 24 |
+
result_selected = pipe(audio)["text"]
|
| 25 |
+
result_other = models[other_model](audio)["text"]
|
| 26 |
+
|
| 27 |
+
return result_selected, result_other
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
title = "ποΈ Moulsot Whisper ASR Comparison"
|
| 31 |
+
description = """
|
| 32 |
+
Compare two fine-tuned Whisper models for **Moroccan ASR**:
|
| 33 |
+
- π© **moulsot_v0.1_2500**
|
| 34 |
+
- π¦ **moulsot_v0.2_1000**
|
| 35 |
+
|
| 36 |
+
You can **record** or **upload** an audio sample, then see transcriptions from both models side by side.
|
| 37 |
+
"""
|
| 38 |
+
|
| 39 |
+
with gr.Blocks(title=title) as demo:
|
| 40 |
+
gr.Markdown(f"# {title}\n{description}")
|
| 41 |
+
|
| 42 |
+
with gr.Row():
|
| 43 |
+
audio_input = gr.Audio(
|
| 44 |
+
sources=["microphone", "upload"],
|
| 45 |
+
type="filepath",
|
| 46 |
+
label="π€ Record or Upload Audio"
|
| 47 |
+
)
|
| 48 |
+
model_choice = gr.Radio(
|
| 49 |
+
["moulsot_v0.1_2500", "moulsot_v0.2_1000"],
|
| 50 |
+
label="Choose Primary Model",
|
| 51 |
+
value="moulsot_v0.1_2500"
|
| 52 |
+
)
|
| 53 |
+
|
| 54 |
+
transcribe_btn = gr.Button("π Transcribe")
|
| 55 |
+
|
| 56 |
+
with gr.Row():
|
| 57 |
+
output_selected = gr.Textbox(label="π© Model 1 Output")
|
| 58 |
+
output_other = gr.Textbox(label="π¦ Model 2 Output")
|
| 59 |
+
|
| 60 |
+
transcribe_btn.click(
|
| 61 |
+
fn=transcribe,
|
| 62 |
+
inputs=[audio_input, model_choice],
|
| 63 |
+
outputs=[output_selected, output_other]
|
| 64 |
+
)
|
| 65 |
+
|
| 66 |
+
# For local testing
|
| 67 |
+
if __name__ == "__main__":
|
| 68 |
+
demo.launch()
|
requirements.txt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
gradio
|
| 2 |
+
transformers
|
| 3 |
+
torch
|