Update app.py

app.py

@@ -14,14 +14,36 @@ from huggingface_hub import hf_hub_download, list_repo_files
 from so_vits_svc_fork.hparams import HParams
 from so_vits_svc_fork.inference.core import Svc
 
+
+###################################################################
+# REPLACE THESE VALUES TO CHANGE THE MODEL REPO/CKPT NAME/SETTINGS
+###################################################################
+# The Hugging Face Hub repo ID - 在这里修改repo_id，可替换成任何已经训练好的模型！
 repo_ids = ["nijisakai/sunyanzi", "kevinwang676/jay"]
+
+# If None, Uses latest ckpt in the repo
 ckpt_name = None
+
+# If None, Uses "kmeans.pt" if it exists in the repo
 cluster_model_name = None
+
+# Set the default f0 type to use - use the one it was trained on.
+# The default for so-vits-svc-fork is "dio".
+# Options: "crepe", "crepe-tiny", "parselmouth", "dio", "harvest"
 default_f0_method = "crepe"
+
+# The default ratio of cluster inference to SVC inference.
+# If cluster_model_name is not found in the repo, this is set to 0.
 default_cluster_infer_ratio = 0.5
+
+# Limit on duration of audio at inference time. increase if you can
+# In this parent app, we set the limit with an env var to 30 seconds
+# If you didnt set env var + you go OOM try changing 9e9 to <=300ish
 duration_limit = int(os.environ.get("MAX_DURATION_SECONDS", 9e9))
+###################################################################
 
 models = []
+speakers = []
 for repo_id in repo_ids:
     # Figure out the latest generator by taking highest value one.
     # Ex. if the repo has: G_0.pth, G_100.pth, G_200.pth, we'd use G_200.pth
@@ -47,7 +69,7 @@ for repo_id in repo_ids:
     generator_path = hf_hub_download(repo_id, ckpt_name)
     config_path = hf_hub_download(repo_id, "config.json")
     hparams = HParams(**json.loads(Path(config_path).read_text()))
-    speakers = list(hparams.spk.keys())
+    speakers.extend(list(hparams.spk.keys()))
     device = "cuda" if torch.cuda.is_available() else "cpu"
     model = Svc(net_g_path=generator_path, config_path=config_path, device=device, cluster_model_path=cluster_model_path)
     models.append(model)