init

README.md

@@ -4,9 +4,16 @@ emoji: 👁
 colorFrom: gray
 colorTo: pink
 sdk: gradio
-sdk_version: 5.1.0
+sdk_version:  4.39.0
 app_file: app.py
 pinned: false
+hf_oauth: true
+hf_oauth_expiration_minutes: 480
+hf_oauth_scopes:
+ - read-repos
+ - write-repos
+ - manage-repos
+ - inference-api
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,85 @@
+import os
+from math import floor
+from typing import Optional
+
+import spaces
+import torch
+import gradio as gr
+from transformers import pipeline
+from transformers.pipelines.audio_utils import ffmpeg_read
+
+# config
+model_name = "kotoba-tech/kotoba-whisper-v2.2"
+example_file = "sample_diarization_japanese.mp3"
+
+# device setting
+if torch.cuda.is_available():
+    torch_dtype = torch.bfloat16
+    device = "cuda"
+    model_kwargs = {'attn_implementation': 'sdpa'}
+else:
+    torch_dtype = torch.float32
+    device = "cpu"
+    model_kwargs = {}
+
+# define the pipeline
+pipe = pipeline(
+    model=model_name,
+    chunk_length_s=15,
+    batch_size=16,
+    torch_dtype=torch_dtype,
+    device=device,
+    model_kwargs=model_kwargs,
+    trust_remote_code=True
+)
+sampling_rate = pipe.feature_extractor.sampling_rate
+
+
+def format_time(start: Optional[float], end: Optional[float]):
+
+    def _format_time(seconds: Optional[float]):
+        if seconds is None:
+            return "[no timestamp available]"
+        minutes = floor(seconds / 60)
+        hours = floor(seconds / 3600)
+        seconds = seconds - hours * 3600 - minutes * 60
+        m_seconds = floor(round(seconds - floor(seconds), 1) * 10)
+        seconds = floor(seconds)
+        return f'{minutes:02}:{seconds:02}.{m_seconds:01}'
+
+    return f"[{_format_time(start)} -> {_format_time(end)}]:"
+
+
+@spaces.GPU
+def get_prediction(inputs):
+    return pipe(inputs, generate_kwargs={"language": "ja", "task": "transcribe"})
+
+
+def transcribe(inputs: str):
+    if inputs is None:
+        raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")
+    with open(inputs, "rb") as f:
+        inputs = f.read()
+    prediction = get_prediction({"array": ffmpeg_read(inputs, sampling_rate), "sampling_rate": sampling_rate})
+    output = ""
+    for n, s in enumerate(prediction["speakers"]):
+        text_timestamped = "\n".join([f"- **{format_time(*c['timestamp'])}** {c['text']}" for c in prediction[f"chunks/{s}"]])
+        output += f'### Speaker {n+1} \n{text_timestamped}\n'
+    return output
+
+
+description = (f"Transcribe and diarize long-form microphone or audio inputs with the click of a button! Demo uses "
+               f"Kotoba-Whisper [{model_name}](https://huggingface.co/{model_name}).")
+title = f"Audio Transcription and Diarization with {os.path.basename(model_name)}"
+shared_config = {"fn": transcribe, "title": title, "description": description, "allow_flagging": "never", "examples": [example_file]}
+o_upload = gr.Markdown()
+o_mic = gr.Markdown()
+i_upload = gr.Interface(
+    inputs=[gr.Audio(sources="upload", type="filepath", label="Audio file")], outputs=gr.Markdown(), **shared_config
+)
+i_mic = gr.Interface(
+    inputs=[gr.Audio(sources="microphone", type="filepath", label="Microphone input")], outputs=gr.Markdown(), **shared_config
+)
+with gr.Blocks() as demo:
+    gr.TabbedInterface([i_upload, i_mic], ["Audio file", "Microphone"])
+demo.queue(api_open=False, default_concurrency_limit=40).launch(show_api=False, show_error=True)

packages.txt

@@ -0,0 +1 @@
+ffmpeg

requirements.txt

@@ -0,0 +1,6 @@
+git+https://github.com/huggingface/transformers
+git+https://github.com/huggingface/diarizers
+torchaudio
+torch
+punctuators==0.0.5
+pyannote.audio