Add private Vevo2 voice studio mode

experiments/vevo2-voice-cloning/run_fm_test.py

@@ -0,0 +1,47 @@
+#!/usr/bin/env python3
+"""Minimal reproducible Vevo2 style-preserving voice-conversion smoke test."""
+
+from __future__ import annotations
+
+import argparse
+import os
+import time
+
+import torch
+
+import models.svc.vevo2.infer_vevo2_fm as vevo
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--source", required=True)
+    parser.add_argument("--reference", required=True)
+    parser.add_argument("--output", required=True)
+    parser.add_argument("--no-pitch-shift", action="store_true")
+    args = parser.parse_args()
+
+    os.makedirs(os.path.dirname(os.path.abspath(args.output)), exist_ok=True)
+    started = time.monotonic()
+    vevo.inference_pipeline = vevo.load_inference_pipeline()
+    loaded = time.monotonic()
+    vevo.vevo2_fm(
+        args.source,
+        args.reference,
+        args.output,
+        shifted_src=not args.no_pitch_shift,
+    )
+    finished = time.monotonic()
+    print(
+        {
+            "model_load_seconds": round(loaded - started, 3),
+            "conversion_seconds": round(finished - loaded, 3),
+            "total_seconds": round(finished - started, 3),
+            "peak_vram_mib": round(torch.cuda.max_memory_allocated() / 1048576, 1),
+            "output": args.output,
+        },
+        flush=True,
+    )
+
+
+if __name__ == "__main__":
+    main()