Upload 3 files

README.md

@@ -5,6 +5,7 @@ colorFrom: indigo
 colorTo: blue
 sdk: gradio
 sdk_version: 5.9.1
+python_version: "3.11"
 app_file: app.py
 pinned: false
 ---
@@ -17,9 +18,10 @@ labeled (Locuteur 1, Locuteur 2, …).
 
 Send base64 audio, get JSON: `{"segments":[{"speaker","start","end"}...], "num_speakers"}`.
 
-**Gated models** — you must accept terms for BOTH on the same account as HF_TOKEN:
+**Gated models** — accept terms for BOTH on the same account as HF_TOKEN:
 - pyannote/speaker-diarization-3.1
 - pyannote/segmentation-3.0
 
-Tuned for 2-person interviews/calls (detected, not hard-forced). Heavier than
-the STT Space — use a paid CPU/GPU tier for usable speed.
+Pinned to Python 3.11 + torch/torchaudio 2.1.2 (last version with
+`torchaudio.AudioMetaData`, which pyannote 3.3.2 requires). Heavier than the STT
+Space — use a paid CPU/GPU tier for usable speed.

app.py

@@ -88,12 +88,24 @@ def diarize_b64(audio_b64):
         os.unlink(tmp_path)
 
         segs = []
-        for turn, _, speaker in diarization.itertracks(yield_label=True):
-            segs.append({
-                "speaker": speaker,            # e.g. "SPEAKER_00"
-                "start": round(float(turn.start), 2),
-                "end": round(float(turn.end), 2),
-            })
+        # pyannote 4.x returns an object whose `.speaker_diarization` yields
+        # (turn, speaker) pairs. pyannote 3.x returns an Annotation with
+        # `.itertracks(yield_label=True)` yielding (turn, _, speaker). Support
+        # both so the Space isn't tied to one pyannote version.
+        if hasattr(diarization, "speaker_diarization"):
+            for turn, speaker in diarization.speaker_diarization:
+                segs.append({
+                    "speaker": str(speaker),
+                    "start": round(float(turn.start), 2),
+                    "end": round(float(turn.end), 2),
+                })
+        else:
+            for turn, _, speaker in diarization.itertracks(yield_label=True):
+                segs.append({
+                    "speaker": str(speaker),       # e.g. "SPEAKER_00"
+                    "start": round(float(turn.start), 2),
+                    "end": round(float(turn.end), 2),
+                })
         segs = _merge_segments(segs)
 
         speakers = sorted({s["speaker"] for s in segs})

requirements.txt

@@ -1,7 +1,6 @@
 gradio==5.9.1
-huggingface_hub>=0.26.0
-pyannote.audio==3.3.2
-torch==2.1.2
-torchaudio==2.1.2
+huggingface_hub>=0.30
+pyannote.audio>=4.0.1
+torchaudio>=2.5.0
 soundfile>=0.12.1
-numpy<2
+numpy>=2.0