From a691f14689a9b4a46eb47167dfbf3e4edd6efc8c Mon Sep 17 00:00:00 2001 From: mush42 Date: Sat, 28 Oct 2023 11:39:59 +0200 Subject: [PATCH] Chunk size and chunk padding are now controlable via commandline arguments --- src/python/piper_train/infer_onnx_streaming.py | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/src/python/piper_train/infer_onnx_streaming.py b/src/python/piper_train/infer_onnx_streaming.py index d33c903..7c1f170 100644 --- a/src/python/piper_train/infer_onnx_streaming.py +++ b/src/python/piper_train/infer_onnx_streaming.py @@ -138,12 +138,27 @@ def main(): parser.add_argument("--noise-scale", type=float, default=0.667) parser.add_argument("--noise-scale-w", type=float, default=0.8) parser.add_argument("--length-scale", type=float, default=1.0) + parser.add_argument( + "--chunk-size", + type=int, + default=45, + help="Number of mel frames to decode at each step" + ) + parser.add_argument( + "--chunk-padding", + type=int, + default=5, + help="Number of mel frames to add to the start and end of the current chunk to reduce decoding artifacts" + ) + args = parser.parse_args() streamer = SpeechStreamer( encoder_path=os.fspath(args.encoder), decoder_path=os.fspath(args.decoder), sample_rate=args.sample_rate, + chunk_size=args.chunk_size, + chunk_padding=args.chunk_padding, ) output_buffer = sys.stdout.buffer