Voice synthesis on ISR

Página 10/30
3 | 4 | 5 | 6 | 7 | 8 | 9 | | 11 | 12 | 13 | 14 | 15

Por ARTRAG

Enlighted (6406)

imagem de ARTRAG

11-06-2016, 00:35

New release with improved accuracy in frequency estimation
Stand alone encoder and ASM files here

Por [WYZ]

Champion (421)

imagem de [WYZ]

11-06-2016, 01:00

Really impressive. Nice work ARTRAG!

Perhaps the PSG volume is relatively high. PSG improvement is the best part.

Por ARTRAG

Enlighted (6406)

imagem de ARTRAG

03-07-2016, 00:39

I did some more tests using 3 channels

This rom will use or the PSG or the SCC but in both cases It uses only 3 channels.
No noise active, only the pitch has been estimated more accurately.

Does it sound better than before?

Use ?usr(n) with n=0..18 from basic, with and without SCC

Por [WYZ]

Champion (421)

imagem de [WYZ]

03-07-2016, 11:19

Indeed the PSG samples are continuous and soft but low frecuencies are "equalized" too high and mask the medium a low freqs.
Really a great step forward.Amazing!

SCC: The speech quality is better than PSG.

IS the sample coder for 8 Harmonics available? I'll like to test them too! Smile

Por PAC

Guardian (5737)

imagem de PAC

03-07-2016, 18:02

Great work Arturo!

Por ARTRAG

Enlighted (6406)

imagem de ARTRAG

03-07-2016, 18:10

I've just uploaded a fixed version of the rom after having fixed a small bug in PSG volumes.

I'm worried that on PSG the result isn't very good unless adding PSG noise.
But I do not find a real general rule to settle its period and to choose at witch channels it has to be applied
SNR does not help eigher, as it is not possible to compute it due to phase distortion.
Shall I keep it as it is :-?

Por ARTRAG

Enlighted (6406)

imagem de ARTRAG

04-07-2016, 00:11

I've released another version of the stand alone voice coder

In order to install it without having Matlab, follow the instructions in the readme.txt

This time the encoder is specialized to use OR the PSG (3 channels) OR the SCC (3 or 5 channels), not both. You have to pass on the command line the number of channels to be used in the encoding. The only two possible values are 3 and 5. In case of incorrect values, the encoder will assume 5 channels.

The wav files have to go in wavs subdir.

The sample code in msxLOFI.asm is able to run the data both with 3 and 5 channels, using the PSG or the SCC (with the PSG, only 3 channels will be played in any case).

No noise added, only more precision in the pitch estimation. Let me know how it goes.
Enjoy

Por ARTRAG

Enlighted (6406)

imagem de ARTRAG

04-07-2016, 00:52

Note that megaroms generated by the sample player are ascii8
Openmsx and bluemsx rom guessers fail to detect the mapper in konami scc

Por santiontanon

Paragon (1027)

imagem de santiontanon

06-07-2016, 01:22

Awesome! finally I was able to run it! (I was struggling trying to find the way to set the romtype in Nekolauncher, but I finally gave up and just ran it from the command line). Really cool Artrag!

I will not have any time during the next couple of weeks, but as soon as I can, I want to play with the Matlab coder Smile

Por ARTRAG

Enlighted (6406)

imagem de ARTRAG

06-07-2016, 09:01

It is interesting that the technique works even with two psgs (6 channels).
There is more than one cartridge with an extra psg.
How can I detect a second psg on msx ?
Witch ports is safe to assume?

Página 10/30
3 | 4 | 5 | 6 | 7 | 8 | 9 | | 11 | 12 | 13 | 14 | 15