Whisper

app.py

@@ -1,3 +1,77 @@
+# import gradio as gr
+
+# gr.Interface.load("models/facebook/fastspeech2-en-ljspeech").launch()
+
+# import gradio as gr
+
+# gr.Interface.load("models/openai/whisper-large-v2").launch()
+
 import gradio as gr
+import torch.cuda
+import whisper
+from whisper.tokenizer import LANGUAGES
+
+gpu = torch.cuda.is_available()
+model = None
+
+DESCRIPTION = """
+<div style="display:flex; gap: 1em; justify-content: center; align-items: center;">
+    <a target="_blank" href="https://github.com/dsymbol">
+        <img alt="GitHub" src="https://img.shields.io/github/followers/dsymbol?style=social">
+    </a>
+    <a target="_blank" href="https://colab.research.google.com/#fileId=https://huggingface.co/spaces/dsymbol/whisper-webui/blob/main/notebook.ipynb">
+        <img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
+    </a>
+    <a target="_blank" href="https://huggingface.co/spaces/dsymbol/whisper-webui" rel="noopener noreferrer"><img
+        src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue" alt="Hugging Face Spaces">
+    </a>
+</div>
+"""
+
+
+def transcribe(recording, file, language, task):
+    if recording and file:
+        text = "Please only use one field."
+    elif not recording and not file:
+        text = "Please use one field."
+    else:
+        language = None if language == "Detect" else language
+        filepath = file if file else recording
+        text = model.transcribe(
+            filepath, task=task.lower(), language=language, fp16=gpu
+        )["text"].strip()
+    return text
+
+
+def interface(model_name="small"):
+    global model
+    model = whisper.load_model(model_name)
+
+    return gr.Interface(
+        fn=transcribe,
+        inputs=[
+            gr.Audio(label="Record", source="microphone", type="filepath"),
+            gr.Audio(label="Upload", source="upload", type="filepath"),
+            gr.Dropdown(
+                label="Language",
+                choices=["Detect"] + sorted([i.title() for i in LANGUAGES.values()]),
+                value="Detect",
+            ),
+            gr.Dropdown(
+                label="Task",
+                choices=["Transcribe", "Translate"],
+                value="Transcribe",
+                info="Whether to perform X->X speech recognition or X->English translation",
+            ),
+        ],
+        outputs=gr.Textbox(label="Transcription", lines=26),
+        theme=gr.themes.Default(),
+        title="Whisper: Transcribe Audio",
+        description=DESCRIPTION,
+        allow_flagging="never",
+    )
+
 
-gr.Interface.load("models/facebook/fastspeech2-en-ljspeech").launch()
+if __name__ == "__main__":
+    demo = interface()
+    demo.queue().launch(debug=True)