Spaces:
Running
on
T4
Running
on
T4
bofenghuang
commited on
Commit
β’
cc46c30
1
Parent(s):
5140605
up
Browse files- run_demo_ct2.py +1 -0
- run_demo_hf.py +26 -20
run_demo_ct2.py
CHANGED
@@ -114,6 +114,7 @@ def maybe_load_cached_pipeline(model_name):
|
|
114 |
# downloaded_model_path = hf_hub_download(repo_id=model_name, filename=CHECKPOINT_FILENAME)
|
115 |
# downloaded_model_path = snapshot_download(repo_id=model_name)
|
116 |
downloaded_model_path = snapshot_download(repo_id=model_name, allow_patterns="ctranslate2/*")
|
|
|
117 |
|
118 |
# model = whisper.load_model(downloaded_model_path, device=device)
|
119 |
model = WhisperModel(downloaded_model_path, device=device, compute_type="float16")
|
|
|
114 |
# downloaded_model_path = hf_hub_download(repo_id=model_name, filename=CHECKPOINT_FILENAME)
|
115 |
# downloaded_model_path = snapshot_download(repo_id=model_name)
|
116 |
downloaded_model_path = snapshot_download(repo_id=model_name, allow_patterns="ctranslate2/*")
|
117 |
+
downloaded_model_path = f"{downloaded_model_path}/ctranslate2"
|
118 |
|
119 |
# model = whisper.load_model(downloaded_model_path, device=device)
|
120 |
model = WhisperModel(downloaded_model_path, device=device, compute_type="float16")
|
run_demo_hf.py
CHANGED
@@ -11,7 +11,11 @@ from transformers.utils.logging import disable_progress_bar
|
|
11 |
warnings.filterwarnings("ignore")
|
12 |
disable_progress_bar()
|
13 |
|
14 |
-
MODEL_NAME = "bofenghuang/whisper-large-v2-cv11-french"
|
|
|
|
|
|
|
|
|
15 |
CHUNK_LENGTH_S = 30
|
16 |
|
17 |
logging.basicConfig(
|
@@ -80,38 +84,38 @@ demo = gr.Blocks()
|
|
80 |
mf_transcribe = gr.Interface(
|
81 |
fn=transcribe,
|
82 |
inputs=[
|
83 |
-
gr.
|
84 |
-
gr.
|
85 |
],
|
86 |
# outputs="text",
|
87 |
-
outputs=gr.
|
88 |
-
layout="horizontal",
|
89 |
theme="huggingface",
|
90 |
title="Whisper French Demo π«π· : Transcribe Audio",
|
91 |
-
description=(
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
),
|
96 |
allow_flagging="never",
|
97 |
)
|
98 |
|
99 |
yt_transcribe = gr.Interface(
|
100 |
fn=yt_transcribe,
|
101 |
-
inputs=[gr.
|
102 |
# outputs=["html", "text"],
|
103 |
outputs=[
|
104 |
-
gr.
|
105 |
-
gr.
|
106 |
],
|
107 |
-
layout="horizontal",
|
108 |
theme="huggingface",
|
109 |
title="Whisper French Demo π«π· : Transcribe YouTube",
|
110 |
-
description=(
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
),
|
115 |
allow_flagging="never",
|
116 |
)
|
117 |
|
@@ -119,4 +123,6 @@ with demo:
|
|
119 |
gr.TabbedInterface([mf_transcribe, yt_transcribe], ["Transcribe Audio", "Transcribe YouTube"])
|
120 |
|
121 |
# demo.launch(server_name="0.0.0.0", debug=True, share=True)
|
122 |
-
demo.launch(enable_queue=True)
|
|
|
|
|
|
11 |
warnings.filterwarnings("ignore")
|
12 |
disable_progress_bar()
|
13 |
|
14 |
+
# MODEL_NAME = "bofenghuang/whisper-large-v2-cv11-french"
|
15 |
+
MODEL_NAME = "bofenghuang/whisper-large-v3-french"
|
16 |
+
# MODEL_NAME = "/home/bhuang/transformers/examples/pytorch/speech-recognition/outputs/hf_whisper/whisper-large-v3-ft-french-pnc-ep5-bs280-lr4e6-wd001-audioaug-specaug"
|
17 |
+
# MODEL_NAME = "/home/bhuang/transformers/examples/pytorch/speech-recognition/outputs/hf_whisper/tmp_model"
|
18 |
+
# MODEL_NAME = "/projects/bhuang/models/asr/public/whisper-large-v3-french"
|
19 |
CHUNK_LENGTH_S = 30
|
20 |
|
21 |
logging.basicConfig(
|
|
|
84 |
mf_transcribe = gr.Interface(
|
85 |
fn=transcribe,
|
86 |
inputs=[
|
87 |
+
gr.components.Audio(sources="microphone", type="filepath", label="Record"),
|
88 |
+
gr.components.Audio(sources="upload", type="filepath", label="Upload File"),
|
89 |
],
|
90 |
# outputs="text",
|
91 |
+
outputs=gr.components.Textbox(label="Transcription", show_copy_button=True),
|
92 |
+
# layout="horizontal",
|
93 |
theme="huggingface",
|
94 |
title="Whisper French Demo π«π· : Transcribe Audio",
|
95 |
+
# description=(
|
96 |
+
# "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the the fine-tuned"
|
97 |
+
# f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and π€ Transformers to transcribe audio files"
|
98 |
+
# " of arbitrary length."
|
99 |
+
# ),
|
100 |
allow_flagging="never",
|
101 |
)
|
102 |
|
103 |
yt_transcribe = gr.Interface(
|
104 |
fn=yt_transcribe,
|
105 |
+
inputs=[gr.components.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL")],
|
106 |
# outputs=["html", "text"],
|
107 |
outputs=[
|
108 |
+
gr.components.HTML(label="YouTube Page"),
|
109 |
+
gr.components.Textbox(label="Transcription", show_copy_button=True),
|
110 |
],
|
111 |
+
# layout="horizontal",
|
112 |
theme="huggingface",
|
113 |
title="Whisper French Demo π«π· : Transcribe YouTube",
|
114 |
+
# description=(
|
115 |
+
# "Transcribe long-form YouTube videos with the click of a button! Demo uses the the fine-tuned checkpoint:"
|
116 |
+
# f" [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and π€ Transformers to transcribe audio files of"
|
117 |
+
# " arbitrary length."
|
118 |
+
# ),
|
119 |
allow_flagging="never",
|
120 |
)
|
121 |
|
|
|
123 |
gr.TabbedInterface([mf_transcribe, yt_transcribe], ["Transcribe Audio", "Transcribe YouTube"])
|
124 |
|
125 |
# demo.launch(server_name="0.0.0.0", debug=True, share=True)
|
126 |
+
# demo.launch(enable_queue=True)
|
127 |
+
# see https://github.com/gradio-app/gradio/issues/2551
|
128 |
+
demo.queue(max_size=10).launch(server_name="0.0.0.0", debug=True, share=True, ssl_certfile="/home/bhuang/tools/cert.pem", ssl_keyfile="/home/bhuang/tools/key.pem", ssl_verify=False)
|