Merge pull request #53 from Googlefan256/main

Build optimization and convert model onnx

Cargo.toml

@@ -6,4 +6,9 @@ members = ["sbv2_api", "sbv2_core", "sbv2_bindings"]
 anyhow = "1.0.86"
 dotenvy = "0.15.7"
 env_logger = "0.11.5"
-ndarray = "0.16.1"
+ndarray = "0.16.1"
+
+[profile.release]
+lto = true
+debug = false
+strip = true

convert/convert_model.py

@@ -126,11 +126,13 @@ torch.onnx.export(
     f"../models/model_{out_name}.onnx",
     verbose=True,
     dynamic_axes={
-        "x_tst": {1: "batch_size"},
+        "x_tst": {0: "batch_size", 1: "x_tst_max_length"},
         "x_tst_lengths": {0: "batch_size"},
-        "tones": {1: "batch_size"},
-        "language": {1: "batch_size"},
-        "bert": {2: "batch_size"},
+        "sid": {0: "batch_size"},
+        "tones": {0: "batch_size", 1: "x_tst_max_length"},
+        "language": {0: "batch_size", 1: "x_tst_max_length"},
+        "bert": {0: "batch_size", 2: "x_tst_max_length"},
+        "style_vec": {0: "batch_size"},
     },
     input_names=[
         "x_tst",

sbv2_bindings/examples/basic.py

@@ -8,11 +8,13 @@ def main():
 
     model.load_sbv2file_from_path("amitaro", "../models/amitaro.sbv2")
     print("All setup is done!")
-    
+
     style_vector = model.get_style_vector("amitaro", 0, 1.0)
     with open("output.wav", "wb") as f:
-        f.write(model.synthesize("おはようございます。", "amitaro", style_vector, 0.0, 0.5))
+        f.write(
+            model.synthesize("おはようございます。", "amitaro", style_vector, 0.0, 0.5)
+        )
 
 
 if __name__ == "__main__":
-    main()
+    main()