Update app.py
Browse files
app.py
CHANGED
|
@@ -1,37 +1,28 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import spaces
|
| 3 |
-
import
|
| 4 |
-
from transformers import pipeline
|
| 5 |
-
import os
|
| 6 |
|
| 7 |
-
#
|
| 8 |
-
#MODEL_ID = "collabora/faster-whisper-large-v2-hindi"
|
| 9 |
MODEL_ID = "collabora/faster-whisper-large-v2-hindi"
|
|
|
|
| 10 |
@spaces.GPU(duration=120)
|
| 11 |
def transcribe_hindi(audio):
|
| 12 |
if audio is None:
|
| 13 |
return "कृपया ऑडियो प्रदान करें।", None
|
| 14 |
|
| 15 |
-
#
|
| 16 |
-
|
| 17 |
-
"automatic-speech-recognition",
|
| 18 |
-
model=MODEL_ID,
|
| 19 |
-
torch_dtype=torch.float16,
|
| 20 |
-
device="cuda",
|
| 21 |
-
chunk_length_s=30,
|
| 22 |
-
batch_size=8
|
| 23 |
-
)
|
| 24 |
|
| 25 |
-
#
|
| 26 |
-
|
| 27 |
-
|
| 28 |
|
| 29 |
# डाउनलोड के लिए .txt फ़ाइल बनाना
|
| 30 |
file_path = "transcription.txt"
|
| 31 |
with open(file_path, "w", encoding="utf-8") as f:
|
| 32 |
-
f.write(
|
| 33 |
|
| 34 |
-
return
|
| 35 |
|
| 36 |
custom_css = """
|
| 37 |
footer {visibility: hidden}
|
|
@@ -40,7 +31,7 @@ footer {visibility: hidden}
|
|
| 40 |
"""
|
| 41 |
|
| 42 |
with gr.Blocks(title="IndicWhisper Collabora GPU") as demo:
|
| 43 |
-
gr.HTML("<div id='header'><h1>🎙️ Hindi Whisper (Collabora
|
| 44 |
|
| 45 |
with gr.Row():
|
| 46 |
with gr.Column():
|
|
@@ -57,7 +48,6 @@ with gr.Blocks(title="IndicWhisper Collabora GPU") as demo:
|
|
| 57 |
lines=10,
|
| 58 |
placeholder="आपका टेक्स्ट यहाँ दिखाई देगा..."
|
| 59 |
)
|
| 60 |
-
# डाउनलोड लिंक के लिए Gradio का File कंपोनेंट
|
| 61 |
download_file = gr.File(
|
| 62 |
label="टैक्स्ट फ़ाइल डाउनलोड करें",
|
| 63 |
visible=True
|
|
@@ -65,7 +55,7 @@ with gr.Blocks(title="IndicWhisper Collabora GPU") as demo:
|
|
| 65 |
|
| 66 |
gr.Markdown("""
|
| 67 |
---
|
| 68 |
-
**सुझाव:** यहाँ से प्राप्त आउटपुट को कॉपी करें
|
| 69 |
""")
|
| 70 |
|
| 71 |
submit_btn.click(
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import spaces
|
| 3 |
+
from faster_whisper import WhisperModel
|
|
|
|
|
|
|
| 4 |
|
| 5 |
+
# कोलाब वाला ही मॉडल आईडी
|
|
|
|
| 6 |
MODEL_ID = "collabora/faster-whisper-large-v2-hindi"
|
| 7 |
+
|
| 8 |
@spaces.GPU(duration=120)
|
| 9 |
def transcribe_hindi(audio):
|
| 10 |
if audio is None:
|
| 11 |
return "कृपया ऑडियो प्रदान करें।", None
|
| 12 |
|
| 13 |
+
# हगिंग फेस के नए CUDA ड्राइवर के लिए compute_type="float16" अनिवार्य है
|
| 14 |
+
model = WhisperModel(MODEL_ID, device="cuda", compute_type="float16")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
+
# व्हिस्पर लार्ज मॉडल के लिए चंकिंग बैकएंड में खुद होती है
|
| 17 |
+
segments, info = model.transcribe(audio, language="hi", beam_size=5)
|
| 18 |
+
full_text = "".join([segment.text for segment in segments])
|
| 19 |
|
| 20 |
# डाउनलोड के लिए .txt फ़ाइल बनाना
|
| 21 |
file_path = "transcription.txt"
|
| 22 |
with open(file_path, "w", encoding="utf-8") as f:
|
| 23 |
+
f.write(full_text.strip())
|
| 24 |
|
| 25 |
+
return full_text.strip(), file_path
|
| 26 |
|
| 27 |
custom_css = """
|
| 28 |
footer {visibility: hidden}
|
|
|
|
| 31 |
"""
|
| 32 |
|
| 33 |
with gr.Blocks(title="IndicWhisper Collabora GPU") as demo:
|
| 34 |
+
gr.HTML("<div id='header'><h1>🎙️ Hindi Whisper (Collabora - faster_whisper)</h1></div>")
|
| 35 |
|
| 36 |
with gr.Row():
|
| 37 |
with gr.Column():
|
|
|
|
| 48 |
lines=10,
|
| 49 |
placeholder="आपका टेक्स्ट यहाँ दिखाई देगा..."
|
| 50 |
)
|
|
|
|
| 51 |
download_file = gr.File(
|
| 52 |
label="टैक्स्ट फ़ाइल डाउनलोड करें",
|
| 53 |
visible=True
|
|
|
|
| 55 |
|
| 56 |
gr.Markdown("""
|
| 57 |
---
|
| 58 |
+
**सुझाव:** यहाँ से प्राप्त आउटपुट को कॉपी करें और अपने **Gemini Gem** में पेस्ट करें ताकि **पञ्चमाक्षर नियमों** (ङ्, ञ्, ण्, न्, म्) के अनुसार शुद्धिकरण किया जा सके।
|
| 59 |
""")
|
| 60 |
|
| 61 |
submit_btn.click(
|