Spaces:
Running
on
Zero
Running
on
Zero
bug_fix env (#47)
Browse files- clean code, requirements compatibility check (524ec453999a3139fc3d9e33dec15a8ac0af6e4f)
- app.py +97 -36
- pyannote/config.yaml +0 -18
- pyannote/pytorch_model.bin +0 -3
- pyannote/segmentation-3.0.bin +0 -3
- pyannote/wespeaker-voxceleb-resnet34-LM.bin +0 -3
- requirements.txt +13 -9
app.py
CHANGED
|
@@ -1,22 +1,38 @@
|
|
| 1 |
import os
|
| 2 |
import gradio as gr
|
| 3 |
import spaces
|
| 4 |
-
|
| 5 |
from whisper_cs_fase_1 import generate_fase_1
|
| 6 |
from whisper_cs_fase_2 import generate_fase_2
|
| 7 |
from AinaTheme import theme
|
| 8 |
|
| 9 |
-
@spaces.GPU()
|
| 10 |
-
def transcribe_fase_1(inputs: str, model_version: str, civil_channel: str):
|
| 11 |
-
if inputs is None:
|
| 12 |
-
raise gr.Error("Cap fitxer d'àudio introduit! Si us plau pengeu un fitxer o enregistreu un àudio abans d'enviar la vostra sol·licitud")
|
| 13 |
-
return generate_fase_1(audio_path=inputs, model_version=model_version, civil_channel=civil_channel)
|
| 14 |
|
| 15 |
-
|
| 16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
if inputs is None:
|
| 18 |
-
raise gr.Error(
|
| 19 |
-
|
|
|
|
|
|
|
| 20 |
|
| 21 |
|
| 22 |
def clear_fase_1(model_version, civil_channel):
|
|
@@ -27,16 +43,17 @@ def clear_fase_2(model_version, civil_channel):
|
|
| 27 |
|
| 28 |
|
| 29 |
with gr.Blocks(theme=theme) as demo:
|
|
|
|
| 30 |
gr.Markdown("## 🗣️ Transcripció automàtica d'àudio — Mode amb dues fases")
|
| 31 |
|
| 32 |
with gr.Tabs():
|
| 33 |
with gr.Tab("Fase 1"):
|
| 34 |
-
|
|
|
|
| 35 |
"### 🎧 Transcripció de trucades multilingüe de bona qualitat per a transcripció fiable\n"
|
| 36 |
"- **v2_fast**: Inclou separació de canals i inferència ràpida.\n"
|
| 37 |
"- **v1.0**: Inclou inferència moderada sense separació de canals."
|
| 38 |
)
|
| 39 |
-
gr.Markdown(description_string)
|
| 40 |
|
| 41 |
with gr.Row():
|
| 42 |
with gr.Column(scale=1):
|
|
@@ -46,28 +63,45 @@ with gr.Blocks(theme=theme) as demo:
|
|
| 46 |
value="v2_fast",
|
| 47 |
elem_id="fase1-model-version",
|
| 48 |
)
|
|
|
|
| 49 |
civil_channel_1 = gr.Dropdown(
|
| 50 |
label="Canal del Civil (persona que truca)",
|
| 51 |
choices=["Left", "Right"],
|
| 52 |
value="Left",
|
| 53 |
)
|
| 54 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 55 |
with gr.Column(scale=1):
|
| 56 |
output_1 = gr.Textbox(label="Output", lines=8)
|
| 57 |
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
-
|
| 63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 64 |
|
| 65 |
with gr.Tab("Fase 2"):
|
| 66 |
-
|
|
|
|
| 67 |
"### 🧠 Transcripció de trucades multilingüe de bona qualitat per a anàlisi d'informe\n"
|
| 68 |
-
"- **v2_fast_and_detection_v1**: Inclou inferència ràpida, separació de parlants i explotació
|
| 69 |
)
|
| 70 |
-
gr.Markdown(description_string)
|
| 71 |
|
| 72 |
with gr.Row():
|
| 73 |
with gr.Column(scale=1):
|
|
@@ -77,33 +111,60 @@ with gr.Blocks(theme=theme) as demo:
|
|
| 77 |
value="v2_fast_and_detection_v1",
|
| 78 |
elem_id="fase2-model-version",
|
| 79 |
)
|
|
|
|
| 80 |
civil_channel_2 = gr.Dropdown(
|
| 81 |
label="Canal del Civil (persona que truca)",
|
| 82 |
choices=["Left", "Right"],
|
| 83 |
value="Left",
|
| 84 |
)
|
| 85 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
with gr.Column(scale=1):
|
| 87 |
output_text = gr.Textbox(label="Transcripció ASR", lines=8)
|
| 88 |
output_sex = gr.Textbox(label="Gènere", lines=1)
|
| 89 |
output_age = gr.Textbox(label="Edat", lines=1)
|
| 90 |
output_silence = gr.Textbox(label="Detecció de silenci", lines=2)
|
| 91 |
output_shout = gr.Textbox(label="Detecció de crits", lines=2)
|
| 92 |
-
output_meteo = gr.Textbox(label="Detecció
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 105 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
|
| 107 |
if __name__ == "__main__":
|
| 108 |
demo.launch()
|
| 109 |
-
|
|
|
|
| 1 |
import os
|
| 2 |
import gradio as gr
|
| 3 |
import spaces
|
|
|
|
| 4 |
from whisper_cs_fase_1 import generate_fase_1
|
| 5 |
from whisper_cs_fase_2 import generate_fase_2
|
| 6 |
from AinaTheme import theme
|
| 7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
+
def generate_fase(audio_path, model_version, civil_channel, fase):
|
| 10 |
+
if fase == 1:
|
| 11 |
+
text = generate_fase_1(
|
| 12 |
+
audio_path,
|
| 13 |
+
model_version=model_version,
|
| 14 |
+
civil_channel=civil_channel
|
| 15 |
+
)
|
| 16 |
+
return text, None, None, None, None, None
|
| 17 |
+
|
| 18 |
+
elif fase == 2:
|
| 19 |
+
text, sex, age, silence_event, shout_event, meteo_event = generate_fase_2(
|
| 20 |
+
audio_path,
|
| 21 |
+
model_version=model_version,
|
| 22 |
+
civil_channel=civil_channel
|
| 23 |
+
)
|
| 24 |
+
return text, sex, age, silence_event, shout_event, meteo_event
|
| 25 |
+
|
| 26 |
+
else:
|
| 27 |
+
raise ValueError("Invalid fase. Must be 1 or 2.")
|
| 28 |
+
|
| 29 |
+
@spaces.GPU
|
| 30 |
+
def transcribe(inputs: str, model_version: str, civil_channel: str, fase: int):
|
| 31 |
if inputs is None:
|
| 32 |
+
raise gr.Error(
|
| 33 |
+
"Cap fitxer d'àudio introduit! Si us plau pengeu un fitxer o enregistreu un àudio abans d'enviar la vostra sol·licitud"
|
| 34 |
+
)
|
| 35 |
+
return generate_fase(inputs, model_version, civil_channel, fase)
|
| 36 |
|
| 37 |
|
| 38 |
def clear_fase_1(model_version, civil_channel):
|
|
|
|
| 43 |
|
| 44 |
|
| 45 |
with gr.Blocks(theme=theme) as demo:
|
| 46 |
+
|
| 47 |
gr.Markdown("## 🗣️ Transcripció automàtica d'àudio — Mode amb dues fases")
|
| 48 |
|
| 49 |
with gr.Tabs():
|
| 50 |
with gr.Tab("Fase 1"):
|
| 51 |
+
|
| 52 |
+
gr.Markdown(
|
| 53 |
"### 🎧 Transcripció de trucades multilingüe de bona qualitat per a transcripció fiable\n"
|
| 54 |
"- **v2_fast**: Inclou separació de canals i inferència ràpida.\n"
|
| 55 |
"- **v1.0**: Inclou inferència moderada sense separació de canals."
|
| 56 |
)
|
|
|
|
| 57 |
|
| 58 |
with gr.Row():
|
| 59 |
with gr.Column(scale=1):
|
|
|
|
| 63 |
value="v2_fast",
|
| 64 |
elem_id="fase1-model-version",
|
| 65 |
)
|
| 66 |
+
|
| 67 |
civil_channel_1 = gr.Dropdown(
|
| 68 |
label="Canal del Civil (persona que truca)",
|
| 69 |
choices=["Left", "Right"],
|
| 70 |
value="Left",
|
| 71 |
)
|
| 72 |
+
|
| 73 |
+
input_1 = gr.Audio(
|
| 74 |
+
sources=["upload", "microphone"],
|
| 75 |
+
type="filepath",
|
| 76 |
+
label="Audio",
|
| 77 |
+
)
|
| 78 |
+
|
| 79 |
with gr.Column(scale=1):
|
| 80 |
output_1 = gr.Textbox(label="Output", lines=8)
|
| 81 |
|
| 82 |
+
with gr.Row(variant="panel"):
|
| 83 |
+
clear_btn = gr.Button("Clear")
|
| 84 |
+
submit_btn = gr.Button("Submit", variant="primary")
|
| 85 |
+
|
| 86 |
+
submit_btn.click(
|
| 87 |
+
fn=transcribe,
|
| 88 |
+
inputs=[input_1, model_version_1, civil_channel_1, gr.Number(value=1)],
|
| 89 |
+
outputs=[output_1],
|
| 90 |
+
)
|
| 91 |
|
| 92 |
+
clear_btn.click(
|
| 93 |
+
fn=clear_fase_1,
|
| 94 |
+
inputs=[model_version_1, civil_channel_1],
|
| 95 |
+
outputs=[input_1, model_version_1, civil_channel_1],
|
| 96 |
+
queue=False,
|
| 97 |
+
)
|
| 98 |
|
| 99 |
with gr.Tab("Fase 2"):
|
| 100 |
+
|
| 101 |
+
gr.Markdown(
|
| 102 |
"### 🧠 Transcripció de trucades multilingüe de bona qualitat per a anàlisi d'informe\n"
|
| 103 |
+
"- **v2_fast_and_detection_v1**: Inclou inferència ràpida, separació de parlants i explotació d'informació detectada."
|
| 104 |
)
|
|
|
|
| 105 |
|
| 106 |
with gr.Row():
|
| 107 |
with gr.Column(scale=1):
|
|
|
|
| 111 |
value="v2_fast_and_detection_v1",
|
| 112 |
elem_id="fase2-model-version",
|
| 113 |
)
|
| 114 |
+
|
| 115 |
civil_channel_2 = gr.Dropdown(
|
| 116 |
label="Canal del Civil (persona que truca)",
|
| 117 |
choices=["Left", "Right"],
|
| 118 |
value="Left",
|
| 119 |
)
|
| 120 |
+
|
| 121 |
+
input_2 = gr.Audio(
|
| 122 |
+
sources=["upload", "microphone"],
|
| 123 |
+
type="filepath",
|
| 124 |
+
label="Audio",
|
| 125 |
+
)
|
| 126 |
+
|
| 127 |
with gr.Column(scale=1):
|
| 128 |
output_text = gr.Textbox(label="Transcripció ASR", lines=8)
|
| 129 |
output_sex = gr.Textbox(label="Gènere", lines=1)
|
| 130 |
output_age = gr.Textbox(label="Edat", lines=1)
|
| 131 |
output_silence = gr.Textbox(label="Detecció de silenci", lines=2)
|
| 132 |
output_shout = gr.Textbox(label="Detecció de crits", lines=2)
|
| 133 |
+
output_meteo = gr.Textbox(label="Detecció meteo", lines=2)
|
| 134 |
+
|
| 135 |
+
with gr.Row(variant="panel"):
|
| 136 |
+
clear_btn2 = gr.Button("Clear")
|
| 137 |
+
submit_btn2 = gr.Button("Submit", variant="primary")
|
| 138 |
+
|
| 139 |
+
submit_btn2.click(
|
| 140 |
+
fn=transcribe,
|
| 141 |
+
inputs=[input_2, model_version_2, civil_channel_2, gr.Number(value=2)],
|
| 142 |
+
outputs=[
|
| 143 |
+
output_text,
|
| 144 |
+
output_sex,
|
| 145 |
+
output_age,
|
| 146 |
+
output_silence,
|
| 147 |
+
output_shout,
|
| 148 |
+
output_meteo,
|
| 149 |
+
],
|
| 150 |
+
)
|
| 151 |
|
| 152 |
+
clear_btn2.click(
|
| 153 |
+
fn=clear_fase_2,
|
| 154 |
+
inputs=[model_version_2, civil_channel_2],
|
| 155 |
+
outputs=[
|
| 156 |
+
input_2,
|
| 157 |
+
model_version_2,
|
| 158 |
+
civil_channel_2,
|
| 159 |
+
output_text,
|
| 160 |
+
output_sex,
|
| 161 |
+
output_age,
|
| 162 |
+
output_silence,
|
| 163 |
+
output_shout,
|
| 164 |
+
output_meteo,
|
| 165 |
+
],
|
| 166 |
+
queue=False,
|
| 167 |
+
)
|
| 168 |
|
| 169 |
if __name__ == "__main__":
|
| 170 |
demo.launch()
|
|
|
pyannote/config.yaml
DELETED
|
@@ -1,18 +0,0 @@
|
|
| 1 |
-
version: 3.1.0
|
| 2 |
-
|
| 3 |
-
pipeline:
|
| 4 |
-
name: pyannote.audio.pipelines.SpeakerDiarization
|
| 5 |
-
params:
|
| 6 |
-
clustering: AgglomerativeClustering
|
| 7 |
-
embedding: ./pyannote/wespeaker-voxceleb-resnet34-LM.bin
|
| 8 |
-
embedding_batch_size: 32
|
| 9 |
-
embedding_exclude_overlap: false
|
| 10 |
-
segmentation: ./pyannote/segmentation-3.0.bin
|
| 11 |
-
segmentation_batch_size: 32
|
| 12 |
-
params:
|
| 13 |
-
clustering:
|
| 14 |
-
method: centroid
|
| 15 |
-
min_cluster_size: 12
|
| 16 |
-
threshold: 0.7045654963945799
|
| 17 |
-
segmentation:
|
| 18 |
-
min_duration_off: 0.09791355693027545
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
pyannote/pytorch_model.bin
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:0b5b3216d60a2d32fc086b47ea8c67589aaeb26b7e07fcbe620d6d0b83e209ea
|
| 3 |
-
size 17719103
|
|
|
|
|
|
|
|
|
|
|
|
pyannote/segmentation-3.0.bin
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:da85c29829d4002daedd676e012936488234d9255e65e86dfab9bec6b1729298
|
| 3 |
-
size 5905440
|
|
|
|
|
|
|
|
|
|
|
|
pyannote/wespeaker-voxceleb-resnet34-LM.bin
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:366edf44f4c80889a3eb7a9d7bdf02c4aede3127f7dd15e274dcdb826b143c56
|
| 3 |
-
size 26645418
|
|
|
|
|
|
|
|
|
|
|
|
requirements.txt
CHANGED
|
@@ -1,16 +1,20 @@
|
|
| 1 |
-
torch
|
| 2 |
-
torchaudio
|
| 3 |
-
|
| 4 |
-
|
|
|
|
| 5 |
faster_whisper==1.2.0
|
|
|
|
|
|
|
| 6 |
hf_transfer==0.1.9
|
| 7 |
-
|
| 8 |
-
yt-dlp==2025.7.21
|
| 9 |
-
gradio==5.41.1
|
| 10 |
librosa==0.10.1
|
|
|
|
| 11 |
ffmpeg-python==0.2.0
|
|
|
|
|
|
|
|
|
|
| 12 |
aina-gradio-theme==2.3
|
| 13 |
spaces==0.39.0
|
| 14 |
-
peft==0.11.1
|
| 15 |
-
whisper_timestamped==1.15.8
|
| 16 |
typing==3.7.4.3
|
|
|
|
|
|
| 1 |
+
torch==2.6.0+cu124 #speechbrain fix
|
| 2 |
+
torchaudio==2.6.0+cu124 #speechbrain fix
|
| 3 |
+
--extra-index-url https://download.pytorch.org/whl/cu124 #speechbrain fix
|
| 4 |
+
nvidia-cudnn-cu12>=9.1.0, <9.2 #ctranslate2 fix
|
| 5 |
+
ctranslate2>=4.3
|
| 6 |
faster_whisper==1.2.0
|
| 7 |
+
whisper_timestamped==1.15.8
|
| 8 |
+
transformers==4.40.2
|
| 9 |
hf_transfer==0.1.9
|
| 10 |
+
huggingface_hub
|
|
|
|
|
|
|
| 11 |
librosa==0.10.1
|
| 12 |
+
soundfile
|
| 13 |
ffmpeg-python==0.2.0
|
| 14 |
+
speechbrain
|
| 15 |
+
pydub
|
| 16 |
+
gradio==5.41.1
|
| 17 |
aina-gradio-theme==2.3
|
| 18 |
spaces==0.39.0
|
|
|
|
|
|
|
| 19 |
typing==3.7.4.3
|
| 20 |
+
yt-dlp==2025.7.21
|