Samples to support our paper titled "SFNet: A computationally efficient source filter model based neural speech synthesis".

This page contains following samples:

1. Original: These are the original recorded speech from target speakers (TS).
2. LPCnet-192: These are synthsized samples using a LPCnet trained using TSP database.
3. SFNet-32: These are synthsized samples from the proposed SF net with fixed gamma tone filterbank with U=32 units
4. SFNet-64: These are synthsized samples from the proposed SF net with fixed gamma tone filterbank with U=64 units
5. SFNet-32-L : These are synthsized samples from the proposed SF net with the learned filterbank with U=32 units

Unseen English speaker samples:

Test file	Original	LPCnet-192	SFNet-32	SFNet-64	SFNet-32-L
M1
M2
F1
F2

Mismatched Language Case Evaluation (Kannada and Hindi):

Test file Original LPCnet-192 SFNet-32 SFNet-64 SFNet-32-L Kannada-1 Kannada-2 Hindi-1 Hindi-2 Thanks for your interest! If you have any questions, please drop us an email: achuthr@iisc.ac.in (Achuth Rao MV), prasantg@iisc.ac.in(Prasanta Kumar Ghosh)