Added the necessary files and scripts for running the app.

README.md

@@ -1,6 +1,6 @@
 ---
-title: Whisper Small Bengali
-emoji: 🐠
+title: Whisper Small Bengali Transcription
+emoji: 💬 🤙
 colorFrom: indigo
 colorTo: yellow
 sdk: gradio

app.py

@@ -0,0 +1,96 @@
+import torch
+from transformers import pipeline
+from transformers.pipelines.audio_utils import ffmpeg_read
+import gradio as gr
+
+MODEL_NAME = "Shamik/whisper-small-bn"
+BATCH_SIZE = 8
+
+device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+
+pipe = pipeline(
+    task="automatic-speech-recognition",
+    model=MODEL_NAME,
+    chunk_length_s=30,
+    device=device,
+)
+
+
+# Copied from https://github.com/openai/whisper/blob/c09a7ae299c4c34c5839a76380ae407e7d785914/whisper/utils.py#L50
+def format_timestamp(seconds: float, always_include_hours: bool = False, decimal_marker: str = "."):
+    if seconds is not None:
+        milliseconds = round(seconds * 1000.0)
+
+        hours = milliseconds // 3_600_000
+        milliseconds -= hours * 3_600_000
+
+        minutes = milliseconds // 60_000
+        milliseconds -= minutes * 60_000
+
+        seconds = milliseconds // 1_000
+        milliseconds -= seconds * 1_000
+
+        hours_marker = f"{hours:02d}:" if always_include_hours or hours > 0 else ""
+        return f"{hours_marker}{minutes:02d}:{seconds:02d}{decimal_marker}{milliseconds:03d}"
+    else:
+        # we have a malformed timestamp so just return it as is
+        return seconds
+
+
+def transcribe(file, task, return_timestamps):
+    outputs = pipe(file, batch_size=BATCH_SIZE, generate_kwargs={"task": "transcribe",
+                                    "language": "bengali"}, return_timestamps=return_timestamps)
+    text = outputs["text"]
+    if return_timestamps:
+        timestamps = outputs["chunks"]
+        timestamps = [
+            f"[{format_timestamp(chunk['timestamp'][0])} -> {format_timestamp(chunk['timestamp'][1])}] {chunk['text']}"
+            for chunk in timestamps
+        ]
+        text = "\n".join(str(feature) for feature in timestamps)
+    return text
+
+demo = gr.Blocks()
+
+mic_transcribe = gr.Interface(
+    fn=transcribe,
+    inputs=[
+        gr.Audio(sources="microphone", type="filepath"),
+        gr.Checkbox(value=False, label="Return timestamps"),
+    ],
+    outputs="text",
+    title="Whisper Bengali Speech Transcription",
+    description=(
+        "Transcribe long-form microphone audio with the click of a button! Demo uses the"
+        f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe audio files"
+        " of arbitrary length."
+    ),
+    allow_flagging="never",
+)
+
+file_transcribe = gr.Interface(
+    fn=transcribe,
+    inputs=[
+        gr.Audio(sources="upload", label="Audio file", type="filepath"),
+        gr.Checkbox(value=False, label="Return timestamps"),
+    ],
+    outputs="text",
+    title="Whisper Bengali Speech Transcription",
+    description=(
+        "Transcribe long-form audio inputs with the click of a button! Demo uses the"
+        f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe audio files"
+        " of arbitrary length."
+    ),
+    examples=[
+        ["./example1.flac",  False],
+        ["./example1.flac",  True],
+    ],
+    cache_examples=True,
+    allow_flagging="never",
+)
+
+with demo:
+    gr.TabbedInterface([mic_transcribe, file_transcribe], ["Transcribe Microphone", "Transcribe Audio File"])
+
+# demo.queue()
+demo.launch()

requirements.txt

@@ -0,0 +1,2 @@
+transformers
+torch