Upload folder using huggingface_hub

ear/semantic.py

@@ -129,12 +129,15 @@ class _Semantic:
                     best_e, best_s = e, s
             mono48 = mono48[best_s : best_s + want]
 
+        clip = mono48.astype(np.float32)
         with self._lock, torch.no_grad():
-            inputs = self._processor(
-                audios=mono48.astype(np.float32),
-                sampling_rate=CLAP_SR,
-                return_tensors="pt",
-            )
+            # transformers 4.x takes `audios`, 5.x renamed it to `audio`.
+            try:
+                inputs = self._processor(audio=clip, sampling_rate=CLAP_SR,
+                                         return_tensors="pt")
+            except TypeError:
+                inputs = self._processor(audios=clip, sampling_rate=CLAP_SR,
+                                         return_tensors="pt")
             audio_emb = self._model.get_audio_features(**inputs)
             audio_emb = audio_emb / audio_emb.norm(dim=-1, keepdim=True)
             sims = (audio_emb @ self._text_emb.T).squeeze(0).cpu().numpy()

requirements.txt

@@ -5,5 +5,7 @@ numpy
 scipy
 soundfile
 torch==2.6.0+cpu
-# ClapProcessor's audio kwarg was renamed in the 5.x line; pin to the 4.x API.
-transformers>=4.44,<5
+# Unpinned on purpose: pinning <5 forces huggingface-hub<1.0, which collides
+# with the version the Spaces runtime preinstalls. semantic.py handles the
+# 4.x/5.x processor signature difference at the call site instead.
+transformers