Voice prompts

Post Reply
VK3KYY
Posts: 7478
Joined: Sat Nov 16, 2019 3:25 am
Location: Melbourne, Australia

Re: Voice prompts

Post by VK3KYY » Mon Jun 08, 2020 11:28 am

OK2HAD wrote:
Mon Jun 08, 2020 11:10 am
VK3KYY wrote:
Mon Jun 08, 2020 9:31 am
OK2HAD wrote:
Mon Jun 08, 2020 8:54 am
Hi, I tried to correct the words in csv for the Czech Republic. I will gradually try Emma, Amy and Nikola. Then I'll let you know which one seems best.
František
OK.

Let me know which voice works best.

It is possible to use phonemes, but the only TTSMP3 supports this, and there is a lower download limit on TTSMP3

Use of Phonemes is documented here

https://docs.aws.amazon.com/polly/lates ... honeme-tag

You can test phonemes on TTSMP3 e..g

<phoneme alphabet="ipa" ph="pɪˈkɑːn">pecan</phoneme>

Do not use the <Speak> tags

The wordlist data format has changed to allow phonemes

But I did not update all the languages yet with the new format.

English has been updated

https://github.com/LibreDMR/voice-promp ... ish_UK.csv

the 2 new columns PromptSpeechPrefix and PromptSpeechPostfix are for phoneme and other XML

e.g.

For <phoneme alphabet="ipa" ph="pɪˈkɑːn">pecan</phoneme>

PromptSpeechPrefix = <phoneme alphabet="ipa" ph="pɪˈkɑːn">
PromptSpeechPostfix = </phoneme>


However I think phonemes may be hard to use, so just changing the spelling of the words to make the speech sound correct is probably easier
Hi, it will be easier for me to edit the words in the text to make them sound believable. On "ttsmp3" I tried a few phrases with different voices. I found the best pronunciation for "Polish / Jan". I will have to adjust the numbers to match the Czech language. Give me a moment and upload a new pronunciation correction file for the pronunciation. It will use Tbin Polish / Jan for the final output of the .bin file. I can't create a .bin file myself. Will it be enough for you if I just edit the text?
OK

Just edit the text, and post it here, I will run my script to create the bin file.

You could create a bin file, but you need to install Python 3 and also FFMPEG.
Also, I have not made a config.csv file for Czech yet, because I did not know which Voice you are using.

BTW. Unless you need to use extra functionality like Phonemes, its probably better to use VoicePolly.pro, because it has a much higher quota.

Both TTSMP3 and VoicePolly both use "Amazon Polly" as for the synthesis, so it should sound the same.

On VoicePolly.pro you can also try setting the sample rate to 8000Hz, because this is the sample rate used by the DMR codec in the radio, so you get a better idea of how it may sound if you select that option.

But the DMR codec also has problems with some voices, especially for individual letters, I know DJ0HF can not make the 2 German female vices work for all letters, but the German male voice (Hans) is better.

VK3KYY
Posts: 7478
Joined: Sat Nov 16, 2019 3:25 am
Location: Melbourne, Australia

Re: Voice prompts

Post by VK3KYY » Mon Jun 08, 2020 11:41 am

Guys

FYI.

I'm going to post an update to all the languages, either later today, or possibly tomorrow, because I have modified my build system to remove the small amount of silence at the start of each voice prompt.

The amount of silence seems to vary, but is usually around 90mS. This may not sound like much, but I may as well remove it, as it makes the voice playback a bit quicker with no loss in quality of audio etc.

But I have to re-encode around 4700 files to AMBE, using the GD-77, and as each file takes around 1 second to encode, the whole process is going to take nearly one and a half hours

So I almost certainly won't be able to get it all done this evening.

VK3KYY
Posts: 7478
Joined: Sat Nov 16, 2019 3:25 am
Location: Melbourne, Australia

Re: Voice prompts

Post by VK3KYY » Mon Jun 08, 2020 12:29 pm

OK.

All languages have been updated to remove the small amount of silence at the start of each number, letter, word or phrase etc

User avatar
Ik0nwg
Posts: 242
Joined: Sat Nov 16, 2019 7:23 am
Location: JN61VG
Contact:

Re: Voice prompts

Post by Ik0nwg » Mon Jun 08, 2020 1:20 pm

VK3KYY wrote:
Mon Jun 08, 2020 12:29 pm
OK.

All languages have been updated to remove the small amount of silence at the start of each number, letter, word or phrase etc
Ciao Roger
In Italian (bianca) there are small problems: example when you pronounce the zero (0) of a frequency you hear a noise at the end of the pronunciation. The voice file before this worked much better.
Thanks for what you do for all of us
sal

User avatar
IU4LEG
Posts: 191
Joined: Wed Nov 20, 2019 12:49 pm

Re: Voice prompts

Post by IU4LEG » Mon Jun 08, 2020 3:07 pm

Ik0nwg wrote:
Mon Jun 08, 2020 1:20 pm
VK3KYY wrote:
Mon Jun 08, 2020 12:29 pm
OK.

All languages have been updated to remove the small amount of silence at the start of each number, letter, word or phrase etc
Ciao Roger
In Italian (bianca) there are small problems: example when you pronounce the zero (0) of a frequency you hear a noise at the end of the pronunciation. The voice file before this worked much better.
Thanks for what you do for all of us
sal
Yes right, i can confirm. Not only with "0" but also with other number when immediately after the number a letter follows, for example "6db", a noise starts between 6 and the letter "d".

User avatar
OK2HAD
Posts: 52
Joined: Wed Jan 08, 2020 1:10 pm
Location: Uherský Brod (Czech Republic)

Re: Voice prompts

Post by OK2HAD » Mon Jun 08, 2020 3:11 pm

VK3KYY wrote:
Mon Jun 08, 2020 11:28 am
OK2HAD wrote:
Mon Jun 08, 2020 11:10 am
VK3KYY wrote:
Mon Jun 08, 2020 9:31 am


OK.

Let me know which voice works best.

It is possible to use phonemes, but the only TTSMP3 supports this, and there is a lower download limit on TTSMP3

Use of Phonemes is documented here

https://docs.aws.amazon.com/polly/lates ... honeme-tag

You can test phonemes on TTSMP3 e..g

<phoneme alphabet="ipa" ph="pɪˈkɑːn">pecan</phoneme>

Do not use the <Speak> tags

The wordlist data format has changed to allow phonemes

But I did not update all the languages yet with the new format.

English has been updated

https://github.com/LibreDMR/voice-promp ... ish_UK.csv

the 2 new columns PromptSpeechPrefix and PromptSpeechPostfix are for phoneme and other XML

e.g.

For <phoneme alphabet="ipa" ph="pɪˈkɑːn">pecan</phoneme>

PromptSpeechPrefix = <phoneme alphabet="ipa" ph="pɪˈkɑːn">
PromptSpeechPostfix = </phoneme>


However I think phonemes may be hard to use, so just changing the spelling of the words to make the speech sound correct is probably easier
Hi, it will be easier for me to edit the words in the text to make them sound believable. On "ttsmp3" I tried a few phrases with different voices. I found the best pronunciation for "Polish / Jan". I will have to adjust the numbers to match the Czech language. Give me a moment and upload a new pronunciation correction file for the pronunciation. It will use Tbin Polish / Jan for the final output of the .bin file. I can't create a .bin file myself. Will it be enough for you if I just edit the text?
OK

Just edit the text, and post it here, I will run my script to create the bin file.

You could create a bin file, but you need to install Python 3 and also FFMPEG.
Also, I have not made a config.csv file for Czech yet, because I did not know which Voice you are using.

BTW. Unless you need to use extra functionality like Phonemes, its probably better to use VoicePolly.pro, because it has a much higher quota.

Both TTSMP3 and VoicePolly both use "Amazon Polly" as for the synthesis, so it should sound the same.

On VoicePolly.pro you can also try setting the sample rate to 8000Hz, because this is the sample rate used by the DMR codec in the radio, so you get a better idea of how it may sound if you select that option.

But the DMR codec also has problems with some voices, especially for individual letters, I know DJ0HF can not make the 2 German female vices work for all letters, but the German male voice (Hans) is better.
Hello,
it worked, but I adjusted the words so that Polish / Jan pronounced correctly and clearly. I checked on Amazon. As soon as I hear the voice in directly in GD-77 I can then do some corrections in the future. Thanks for your work
Attachments
czech_voice2.zip
(2.52 KiB) Downloaded 129 times

VK3KYY
Posts: 7478
Joined: Sat Nov 16, 2019 3:25 am
Location: Melbourne, Australia

Re: Voice prompts

Post by VK3KYY » Mon Jun 08, 2020 9:38 pm

IU4LEG wrote:
Mon Jun 08, 2020 3:07 pm
Ik0nwg wrote:
Mon Jun 08, 2020 1:20 pm
VK3KYY wrote:
Mon Jun 08, 2020 12:29 pm
OK.

All languages have been updated to remove the small amount of silence at the start of each number, letter, word or phrase etc
Ciao Roger
In Italian (bianca) there are small problems: example when you pronounce the zero (0) of a frequency you hear a noise at the end of the pronunciation. The voice file before this worked much better.
Thanks for what you do for all of us
sal
Ok.

Yes right, i can confirm. Not only with "0" but also with other number when immediately after the number a letter follows, for example "6db", a noise starts between 6 and the letter "d".
Ok.

I will investigate why this caused a problem

VK3KYY
Posts: 7478
Joined: Sat Nov 16, 2019 3:25 am
Location: Melbourne, Australia

Re: Voice prompts

Post by VK3KYY » Tue Jun 09, 2020 1:26 am

I found a bug in the encoder, where the end of the audio could contain the data from the previous 60mS audio, if the final "frame" of audio was less than 60mS.

However, even with this bug fixed, I think possibly for some voices, that it is better to have a small gap between each letter, otherwise the speechstill sounds incorrect.

So I'm only going to remove the silence at the beginning of the audio, for languages which I can test myself e.g. English ...

I'll re-generate the files for all other languages, and I'll also double check the English versions are OK with the silence removed.

User avatar
Ik0nwg
Posts: 242
Joined: Sat Nov 16, 2019 7:23 am
Location: JN61VG
Contact:

Re: Voice prompts

Post by Ik0nwg » Tue Jun 09, 2020 1:46 am

VK3KYY wrote:
Tue Jun 09, 2020 1:26 am
I found a bug in the encoder, where the end of the audio could contain the data from the previous 60mS audio, if the final "frame" of audio was less than 60mS.

However, even with this bug fixed, I think possibly for some voices, that it is better to have a small gap between each letter, otherwise the speechstill sounds incorrect.

So I'm only going to remove the silence at the beginning of the audio, for languages which I can test myself e.g. English ...

I'll re-generate the files for all other languages, and I'll also double check the English versions are OK with the silence removed.
Roger
i am using the english version and it works perfectly does not have the problem that is generated like the italian one
Ciao
sal

VK3KYY
Posts: 7478
Joined: Sat Nov 16, 2019 3:25 am
Location: Melbourne, Australia

Re: Voice prompts

Post by VK3KYY » Tue Jun 09, 2020 1:50 am

Yes.

I regenerated them all with the initial silence removed.

It seems that the Italian voices need some space between each letter or number..

Probably the AMBE codec was designed for English :-(

Post Reply