From 9a8a8e3195359144ab510fa75df75e50a2ef4bdc Mon Sep 17 00:00:00 2001 From: mrq Date: Tue, 18 Mar 2025 08:40:43 -0500 Subject: [PATCH] off by one bateman --- vall_e/models/base.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/vall_e/models/base.py b/vall_e/models/base.py index 027f99a..280b659 100755 --- a/vall_e/models/base.py +++ b/vall_e/models/base.py @@ -167,7 +167,7 @@ class AudioEmbedding(nn.Module): # sum all prior codebook levels if requested (as quant_level = 0 does not have any other codebooks to sum through) if sums and quant_level > 0: - x = sum( [ self.embeddings[input_quant_level + offset]( xi[:, input_quant_level] ) for input_quant_level in range( quant_level ) ] ) + x = sum( [ self.embeddings[input_quant_level + offset]( xi[:, input_quant_level] ) for input_quant_level in range( quant_level + 1 ) ] ) else: input_quant_level = quant_level x = self.embeddings[input_quant_level + offset]( xi if xi.dim() == 1 else xi[:, input_quant_level] ) @@ -1719,4 +1719,4 @@ if __name__ == "__main__": resp = generate( phn, prom, sequence=resp, mode=f"resp|NAR:{i-1}:{i}" ) print( f"NAR:{i-1}:{i}: ", resp[-1] ) - decode_to_file( torch.tensor(resp, dtype=torch.int16, device=device).t(), "./data/test.wav" ) \ No newline at end of file + decode_to_file( torch.tensor(resp, dtype=torch.int16, device=device).t(), "./data/test.wav" )