Voice synthesis on ISR

Page 30/31
23 | 24 | 25 | 26 | 27 | 28 | 29 | | 31

By Giangiacomo Zaffini 2

Master (248)

Giangiacomo Zaffini 2's picture

23-12-2019, 14:28

@ [WYZ] impressive! I Your software development achieves true fulfillment of SCC/SCC+ chip capability premises. Cool

By [WYZ]

Champion (428)

[WYZ]'s picture

24-12-2019, 00:12

Thank you Giangiacomo.

File: TRUE SURVIVOR MSX1 finally is a 64kb MegaROM Konami5 SCC. Works with MegaFlashROM+SD using Internal PSG and External PSG + SCC

By Manuel

Ascended (17692)

Manuel's picture

24-12-2019, 00:26

OFFTOPIC: I keep being intrigued by people using the name "Konami5".... (why oh why... I guess it's Marat's fault Tongue)

By [WYZ]

Champion (428)

[WYZ]'s picture

24-12-2019, 00:30

By ARTRAG

Enlighted (6493)

ARTRAG's picture

25-12-2019, 01:09

By alexito

Paladin (698)

alexito's picture

16-02-2020, 23:38

@ARTRAG&WYZ: Simply AMAZING ! Sorry guys for be a little distracted with others topics but I'm very happy for the results achieved in this matter (PCM Samples) I can see a big evolution since the last time I was reading this by 2016. BA-team

By ARTRAG

Enlighted (6493)

ARTRAG's picture

17-02-2020, 18:59

thanks alexito, the latest player works both in the frequency domain and in the time domain
so it isn't properly a PCM player and it can play a voice per scc channel
I should have the time and the energy to release a polished version of encoder and player

By Grauw

Ascended (9687)

Grauw's picture

06-03-2021, 21:03

Still very cool.

About the whistling noise, am I correct that these are mismatches in the base frequency detection? Could they be reduced by setting a frequency band curve to help the algorithm weigh the possible matches? And then for more precision, varying it over time matching the melody? Possibly automatically tracking, assuming that vocals don’t instantly change pitch?

Also I wonder if software like Izotope Melodyne could play a role here.

By ARTRAG

Enlighted (6493)

ARTRAG's picture

07-03-2021, 11:15

I did the same guess about the whistling noise
I was thinking it was due to jumps in the pitch detection among frames
But actually the pitch seems quite continuous when plotted.
Look at this picture I've include din the git repository of the encoder
The pitch is the blue curve (it is in red when we face an unvoiced segment)
https://github.com/artrag/voicenc_scc/blob/master/pitch_catt...

My other guess is that the noise is due to discontinuities between adjacent frames
This is an example of what you see inspecting the result.
Look halfway between 0,51 and 0,52
https://github.com/artrag/voicenc_scc/blob/master/noise_catt...

The solution I was exploring was to "rotate" the phase of the 32 samples to get as first sample the closest one to the last in the previous frame. But early tests were disappointing.

By ARTRAG

Enlighted (6493)

ARTRAG's picture

07-03-2021, 11:42

I have uploaded the latest encoder on github
Look at tt_voicenc_scc.m
If you have matlab (you need also voicebox) and you want to experiment any suggestion is welcome

Page 30/31
23 | 24 | 25 | 26 | 27 | 28 | 29 | | 31