Spaces:
Paused
Paused
Malina commited on
Commit Β·
b1fe369
1
Parent(s): b0258be
NINA v0.0 - video transcription
Browse files- main.py +115 -0
- packages.txt +1 -0
- requirements.txt +4 -0
main.py
ADDED
|
@@ -0,0 +1,115 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 2 |
+
# Neural Information & Narrative Abstraction
|
| 3 |
+
# ver 0.00
|
| 4 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 5 |
+
|
| 6 |
+
import os
|
| 7 |
+
import tempfile
|
| 8 |
+
|
| 9 |
+
import ffmpeg # audio extraction
|
| 10 |
+
import whisper # ASR / speech-to-text
|
| 11 |
+
import gradio as gr # UI + HF Spaces runtime
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
# ββ Model load ββββββββββββββββββββββββββββββ
|
| 15 |
+
print("[NINA] Loading model...")
|
| 16 |
+
asr_model = whisper.load_model("medium") # swap to "small" or "medium" for better accuracy
|
| 17 |
+
print("[NINA] Ready.")
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
# ββ Stage 1: extract audio from video βββββββββββββββββββββββββββββββββββββββββ
|
| 21 |
+
def extract_audio(video_path: str) -> str:
|
| 22 |
+
"""
|
| 23 |
+
Pull the audio track from an mp4, write it as a
|
| 24 |
+
16kHz mono WAV to a temp file, return the path.
|
| 25 |
+
"""
|
| 26 |
+
tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
|
| 27 |
+
tmp.close()
|
| 28 |
+
|
| 29 |
+
(
|
| 30 |
+
ffmpeg
|
| 31 |
+
.input(video_path)
|
| 32 |
+
.output(
|
| 33 |
+
tmp.name,
|
| 34 |
+
ac=1,
|
| 35 |
+
ar=16000,
|
| 36 |
+
vn=None,
|
| 37 |
+
)
|
| 38 |
+
.run(overwrite_output=True, quiet=True)
|
| 39 |
+
)
|
| 40 |
+
|
| 41 |
+
return tmp.name
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
# ββ Stage 2: transcribe audio β plain text ββββββββββββββββββββββββββββββββββββ
|
| 45 |
+
def transcribe_audio(audio_path: str) -> list[dict]:
|
| 46 |
+
"""
|
| 47 |
+
Run Whisper on the audio file.
|
| 48 |
+
Returns a list of segment dicts:
|
| 49 |
+
{ text, start, end }
|
| 50 |
+
"""
|
| 51 |
+
result = asr_model.transcribe(
|
| 52 |
+
audio_path,
|
| 53 |
+
word_timestamps=False,
|
| 54 |
+
verbose=False,
|
| 55 |
+
)
|
| 56 |
+
|
| 57 |
+
segments = [
|
| 58 |
+
{
|
| 59 |
+
"text": seg["text"].strip(),
|
| 60 |
+
"start": round(seg["start"], 2),
|
| 61 |
+
"end": round(seg["end"], 2),
|
| 62 |
+
}
|
| 63 |
+
for seg in result["segments"]
|
| 64 |
+
]
|
| 65 |
+
|
| 66 |
+
return segments
|
| 67 |
+
|
| 68 |
+
|
| 69 |
+
# ββ Formatter: segments β readable transcript string ββββββββββββββββββββββββββ
|
| 70 |
+
def format_transcript(segments: list[dict]) -> str:
|
| 71 |
+
lines = []
|
| 72 |
+
for seg in segments:
|
| 73 |
+
timestamp = f"[{seg['start']:>7.2f}s β {seg['end']:>7.2f}s]"
|
| 74 |
+
lines.append(f"{timestamp} {seg['text']}")
|
| 75 |
+
return "\n".join(lines)
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
# ββ Main pipeline (v0) ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 79 |
+
def run(video_path: str) -> str:
|
| 80 |
+
if video_path is None:
|
| 81 |
+
return " No video uploaded."
|
| 82 |
+
|
| 83 |
+
print(f"[NINA] Processing: {video_path}")
|
| 84 |
+
|
| 85 |
+
# 1. extract audio
|
| 86 |
+
audio_path = extract_audio(video_path)
|
| 87 |
+
print(f"[NINA] Audio extracted β {audio_path}")
|
| 88 |
+
|
| 89 |
+
# 2. transcribe
|
| 90 |
+
segments = transcribe_audio(audio_path)
|
| 91 |
+
print(f"[NINA] Transcribed {len(segments)} segments.")
|
| 92 |
+
|
| 93 |
+
# 3. cleanup temp audio file
|
| 94 |
+
os.remove(audio_path)
|
| 95 |
+
|
| 96 |
+
# 4. format output
|
| 97 |
+
transcript = format_transcript(segments)
|
| 98 |
+
return transcript
|
| 99 |
+
|
| 100 |
+
|
| 101 |
+
# ββ Gradio UI βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 102 |
+
demo = gr.Interface(
|
| 103 |
+
fn=run,
|
| 104 |
+
inputs=gr.Video(label="Upload video"),
|
| 105 |
+
outputs=gr.Textbox(
|
| 106 |
+
label="Transcription",
|
| 107 |
+
lines=25,
|
| 108 |
+
),
|
| 109 |
+
title="Neural Information & Narrative Abstraction v0.00 ",
|
| 110 |
+
description="Upload an MP4. Get a full timestamped transcript.",
|
| 111 |
+
flagging_mode="never",
|
| 112 |
+
)
|
| 113 |
+
|
| 114 |
+
if __name__ == "__main__":
|
| 115 |
+
demo.launch()
|
packages.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
ffmpeg
|
requirements.txt
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
ffmpeg-python
|
| 2 |
+
openai-whisper
|
| 3 |
+
gradio
|
| 4 |
+
|