Try OpenAI Whisper

.python-version

@@ -0,0 +1 @@
+3.9.9

README.md

@@ -6,6 +6,7 @@ colorTo: indigo
 sdk: gradio
 sdk_version: 5.30.0
 app_file: app.py
+python_version: 3.9.9
 pinned: false
 license: unlicense
 short_description: Minimum working OpenAI Whisper pipeline in just 19 lines.

app.py

@@ -1,21 +1,14 @@
 """Minimum code to run whisper-faster on Hugging Face"""
-from functools import lru_cache
+import whisper
 import gradio as gr
 import spaces
-from faster_whisper import WhisperModel
-
-@lru_cache(maxsize=1)
-def get_model():
-    """Model caching function"""
-    return WhisperModel("large-v3", device="cuda", compute_type="float16")
-
 
 @spaces.GPU(duration=120)
 def transcribe(gradio_input_one_return_audio_filepath: str):
     """Makes a call to transcribe the audio"""
-    model = get_model()
-    segments, _ = model.transcribe(gradio_input_one_return_audio_filepath)
-    return "\n".join([seg.text for seg in segments])
+    model = whisper.load_model("base").to("cuda")
+    res = model.transcribe(gradio_input_one_return_audio_filepath)
+    return res["text"]
 
 interface = gr.Interface(
     fn = transcribe,

apt.txt

@@ -1,3 +1 @@
 ffmpeg
-libsndfile1
-

requirements.txt

@@ -1,5 +1,3 @@
-gradio>=5.30.0
-faster-whisper
-ctranslate2
-ffmpeg-python
+gradio
 spaces
+openai-whisper