Upload app.py with huggingface_hub

app.py

@@ -1,10 +1,12 @@
+import io
 import os
 import gradio as gr
 import numpy as np
 import pandas as pd
+import soundfile as sf
 import torch
 import torchaudio
-from datasets import load_dataset
+from datasets import load_dataset, Audio
 from sklearn.cluster import AgglomerativeClustering
 from transformers import Wav2Vec2FeatureExtractor, WavLMForXVector
 
@@ -63,12 +65,15 @@ def identify_speakers(
         progress((i + 0.5) / len(repos), desc=f"[{i+1}/{len(repos)}] {short}")
         try:
             ds = load_dataset(repo, split="train", streaming=True, token=token)
+            ds = ds.cast_column("audio", Audio(decode=False))
             embs = []
             for j, row in enumerate(ds):
                 if j >= int(samples_per_book):
                     break
-                audio = row["audio"]
-                embs.append(_embed(np.array(audio["array"]), audio["sampling_rate"]))
+                raw = row["audio"]
+                audio_bytes = raw.get("bytes") or open(raw["path"], "rb").read()
+                audio_array, sr = sf.read(io.BytesIO(audio_bytes))
+                embs.append(_embed(audio_array, sr))
             if embs:
                 embeddings[repo] = np.mean(embs, axis=0)
             else: