Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -7,19 +7,18 @@ from transformers.pipelines.audio_utils import ffmpeg_read
|
|
| 7 |
# -----------------------------
|
| 8 |
# CONFIG
|
| 9 |
# -----------------------------
|
| 10 |
-
ASR_MODEL = "openai/whisper-large-v3"
|
| 11 |
BATCH_SIZE = 8
|
| 12 |
HAS_CUDA = torch.cuda.is_available()
|
| 13 |
DEVICE = 0 if HAS_CUDA else "cpu"
|
| 14 |
DTYPE = torch.float16 if HAS_CUDA else torch.float32
|
| 15 |
|
| 16 |
-
# โหลดโมเดลครั้งเดียว
|
| 17 |
asr_pipe = pipeline(
|
| 18 |
task="automatic-speech-recognition",
|
| 19 |
model=ASR_MODEL,
|
| 20 |
device=DEVICE,
|
| 21 |
torch_dtype=DTYPE,
|
| 22 |
-
chunk_length_s=30,
|
| 23 |
)
|
| 24 |
|
| 25 |
def _save_text_file(text: str, suffix: str = ".txt") -> str:
|
|
@@ -28,10 +27,34 @@ def _save_text_file(text: str, suffix: str = ".txt") -> str:
|
|
| 28 |
f.write(text)
|
| 29 |
return path
|
| 30 |
|
| 31 |
-
def
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
with open(path, "rb") as f:
|
| 36 |
payload = f.read()
|
| 37 |
audio = ffmpeg_read(payload, asr_pipe.feature_extractor.sampling_rate)
|
|
@@ -40,22 +63,21 @@ def _transcribe_from_path(path: str, task: str):
|
|
| 40 |
out = asr_pipe(
|
| 41 |
inputs,
|
| 42 |
batch_size=BATCH_SIZE,
|
| 43 |
-
generate_kwargs={"task": task}, # 'transcribe' =
|
| 44 |
return_timestamps=True,
|
| 45 |
)
|
| 46 |
text = out["text"]
|
| 47 |
return text, _save_text_file(text, ".txt")
|
| 48 |
|
| 49 |
-
# ---- entry points สำหรับ UI สามแท็บ ----
|
| 50 |
def transcribe_mic(mic_path: str, task: str):
|
| 51 |
-
return
|
| 52 |
|
| 53 |
def transcribe_audio(file_path: str, task: str):
|
| 54 |
-
return
|
| 55 |
|
| 56 |
-
def transcribe_video(
|
| 57 |
-
#
|
| 58 |
-
return
|
| 59 |
|
| 60 |
# -----------------------------
|
| 61 |
# UI
|
|
@@ -78,7 +100,8 @@ with gr.Blocks(title="Whisper V3 – Transcriber (Audio + MP4)") as demo:
|
|
| 78 |
gr.Button("Run").click(transcribe_audio, inputs=[up_audio, up_task], outputs=[up_text, up_file])
|
| 79 |
|
| 80 |
with gr.Tab("🎬 Video MP4"):
|
| 81 |
-
|
|
|
|
| 82 |
vd_task = gr.Radio(["transcribe", "translate"], value="transcribe", label="Task")
|
| 83 |
vd_text = gr.Textbox(label="Transcript", lines=10)
|
| 84 |
vd_file = gr.File(label="Download Transcript (.txt)")
|
|
|
|
| 7 |
# -----------------------------
|
| 8 |
# CONFIG
|
| 9 |
# -----------------------------
|
| 10 |
+
ASR_MODEL = "openai/whisper-large-v3"
|
| 11 |
BATCH_SIZE = 8
|
| 12 |
HAS_CUDA = torch.cuda.is_available()
|
| 13 |
DEVICE = 0 if HAS_CUDA else "cpu"
|
| 14 |
DTYPE = torch.float16 if HAS_CUDA else torch.float32
|
| 15 |
|
|
|
|
| 16 |
asr_pipe = pipeline(
|
| 17 |
task="automatic-speech-recognition",
|
| 18 |
model=ASR_MODEL,
|
| 19 |
device=DEVICE,
|
| 20 |
torch_dtype=DTYPE,
|
| 21 |
+
chunk_length_s=30,
|
| 22 |
)
|
| 23 |
|
| 24 |
def _save_text_file(text: str, suffix: str = ".txt") -> str:
|
|
|
|
| 27 |
f.write(text)
|
| 28 |
return path
|
| 29 |
|
| 30 |
+
def _resolve_path(x):
|
| 31 |
+
"""
|
| 32 |
+
รับค่า input ได้ทั้ง:
|
| 33 |
+
- str (filepath)
|
| 34 |
+
- dict ที่มี key 'name' หรือ 'path'
|
| 35 |
+
- gradio FileData (มี .path)
|
| 36 |
+
คืนค่าเป็น filepath เสมอ
|
| 37 |
+
"""
|
| 38 |
+
if x is None:
|
| 39 |
+
return None
|
| 40 |
+
if isinstance(x, str):
|
| 41 |
+
return x
|
| 42 |
+
# Gradio v4/v5 บางทีให้เป็น dict
|
| 43 |
+
if isinstance(x, dict):
|
| 44 |
+
return x.get("path") or x.get("name")
|
| 45 |
+
# Gradio FileData
|
| 46 |
+
path = getattr(x, "path", None)
|
| 47 |
+
if path:
|
| 48 |
+
return path
|
| 49 |
+
# เผื่อกรณี edge
|
| 50 |
+
return str(x)
|
| 51 |
+
|
| 52 |
+
def _transcribe_from_any(file_like, task: str):
|
| 53 |
+
path = _resolve_path(file_like)
|
| 54 |
+
if not path or not os.path.exists(path):
|
| 55 |
+
raise gr.Error("ไม่พบไฟล์ที่อัปโหลด (path ว่างหรือไฟล์หาย)")
|
| 56 |
+
|
| 57 |
+
# อ่านเป็น bytes แล้วให้ ffmpeg แปลงเป็นโมโน float32
|
| 58 |
with open(path, "rb") as f:
|
| 59 |
payload = f.read()
|
| 60 |
audio = ffmpeg_read(payload, asr_pipe.feature_extractor.sampling_rate)
|
|
|
|
| 63 |
out = asr_pipe(
|
| 64 |
inputs,
|
| 65 |
batch_size=BATCH_SIZE,
|
| 66 |
+
generate_kwargs={"task": task}, # 'transcribe' = ตามภาษาเดิม, 'translate' = แปลอังกฤษ
|
| 67 |
return_timestamps=True,
|
| 68 |
)
|
| 69 |
text = out["text"]
|
| 70 |
return text, _save_text_file(text, ".txt")
|
| 71 |
|
|
|
|
| 72 |
def transcribe_mic(mic_path: str, task: str):
|
| 73 |
+
return _transcribe_from_any(mic_path, task)
|
| 74 |
|
| 75 |
def transcribe_audio(file_path: str, task: str):
|
| 76 |
+
return _transcribe_from_any(file_path, task)
|
| 77 |
|
| 78 |
+
def transcribe_video(video_file, task: str):
|
| 79 |
+
# รับเป็นไฟล์วิดีโอ (mp4, mov, webm ก็ได้) แล้วให้ ffmpeg_read ดึงเสียง
|
| 80 |
+
return _transcribe_from_any(video_file, task)
|
| 81 |
|
| 82 |
# -----------------------------
|
| 83 |
# UI
|
|
|
|
| 100 |
gr.Button("Run").click(transcribe_audio, inputs=[up_audio, up_task], outputs=[up_text, up_file])
|
| 101 |
|
| 102 |
with gr.Tab("🎬 Video MP4"):
|
| 103 |
+
# ใช้ gr.File เพื่อให้ได้ path ที่นิ่งที่สุด
|
| 104 |
+
up_video = gr.File(file_count="single", file_types=["video"], label="Upload MP4 / Video")
|
| 105 |
vd_task = gr.Radio(["transcribe", "translate"], value="transcribe", label="Task")
|
| 106 |
vd_text = gr.Textbox(label="Transcript", lines=10)
|
| 107 |
vd_file = gr.File(label="Download Transcript (.txt)")
|