Building voice prompts

[VK3KYY]

To build the voice prompts you need to download the build tools

downloads/Voice_Prompts/tools



You also need Python installed on your computer

On windows you can use the build_all.bat file to do a complete build.

Note. You must have an opengd77 radio connected to your PC and turned on, because the scripts need to use the radio to compress the data into AMBE format. AMBE has IP issues and hence AFIK there is no AMBE compressor program that runs natively on either Windows or Linux or Mac etc

[KK7NTZ]

The data folder redirects to 404, did it get deleted?

Wayback machine has one entry from 2022 and shows a file called languages.zip that isn't archived: https://web.archive.org/web/20221207164 ... mpts/data/

73,
Jayden - KK7NTZ

[VK3KYY]

The data folder redirects to 404, did it get deleted?

Wayback machine has one entry from 2022 and shows a file called languages.zip that isn't archived: https://web.archive.org/web/20221207164 ... mpts/data/

73,
Jayden - KK7NTZ

Insufficient server space

You now need to build from scratch

[UB3GDZ]

Hello everyone ! Maybe someone has done something like this, I got it all right and made Russian prompts, but in Russian the prompts do not sound as good as I would like, accents are not where they should be, as I did not try to edit the spelling ! I tried a crazy variant, which I got not as I expected ! I found on the net a service with AI, which pronounces words very well, I wrote each word manually on this service and downloaded them to PC giving them names like the necessary files and then running the script I get ready voice packets, but the packet with slow pronunciation is not collected due to exceeding the limit of 166912 bytes, and the other packets are too fast pronounced ! Is there no way to compress the source files a bit ? Just after converting .mp3 to .RAW, the file is twice as big as the original ! And my package starts to have a volume of 191000 bytes ! I know this sounds like madness but it's just tests and maybe there is a solution !!!! Thanks ...

[VK3KYY]

Hello everyone ! Maybe someone has done something like this, I got it all right and made Russian prompts, but in Russian the prompts do not sound as good as I would like, accents are not where they should be, as I did not try to edit the spelling ! I tried a crazy variant, which I got not as I expected ! I found on the net a service with AI, which pronounces words very well, I wrote each word manually on this service and downloaded them to PC giving them names like the necessary files and then running the script I get ready voice packets, but the packet with slow pronunciation is not collected due to exceeding the limit of 166912 bytes, and the other packets are too fast pronounced ! Is there no way to compress the source files a bit ? Just after converting .mp3 to .RAW, the file is twice as big as the original ! And my package starts to have a volume of 191000 bytes ! I know this sounds like madness but it's just tests and maybe there is a solution !!!! Thanks ...

The size of the RAW file will only be dependent on the length of the audio. Assuming you didn't change the sample rate or other factors of the mp3

The mp3 files created by ttsmp3.com are 22050Hz sample rate, which give bitrate of 48kb/s

The RAW files are 16 bit signged PCM, mono , 8000Hz sample rate

Check the data format of your RAW files using Audacity etc to confirm that they are the correct format

BTW.

We can't accept manually created vpr files, because its not possible for other people to replicate the files themselves.

We use ttsmp3.com becasue its a publically available website and is free and does not require a login etc

The whole process is automated and only requires the user to install ffmpeg and python, which are both publically available for multiple platforms

[UB3GDZ]

Hello everyone ! Maybe someone has done something like this, I got it all right and made Russian prompts, but in Russian the prompts do not sound as good as I would like, accents are not where they should be, as I did not try to edit the spelling ! I tried a crazy variant, which I got not as I expected ! I found on the net a service with AI, which pronounces words very well, I wrote each word manually on this service and downloaded them to PC giving them names like the necessary files and then running the script I get ready voice packets, but the packet with slow pronunciation is not collected due to exceeding the limit of 166912 bytes, and the other packets are too fast pronounced ! Is there no way to compress the source files a bit ? Just after converting .mp3 to .RAW, the file is twice as big as the original ! And my package starts to have a volume of 191000 bytes ! I know this sounds like madness but it's just tests and maybe there is a solution !!!! Thanks ...

The size of the RAW file will only be dependent on the length of the audio. Assuming you didn't change the sample rate or other factors of the mp3

The mp3 files created by ttsmp3.com are 22050Hz sample rate, which give bitrate of 48kb/s

The RAW files are 16 bit signged PCM, mono , 8000Hz sample rate

Check the data format of your RAW files using Audacity etc to confirm that they are the correct format

BTW.

We can't accept manually created vpr files, because its not possible for other people to replicate the files themselves.

We use ttsmp3.com becasue its a publically available website and is free and does not require a login etc

The whole process is automated and only requires the user to install ffmpeg and python, which are both publically available for multiple platforms

Thanks for the reply ! I realise that I am doing this as a test and experiment for a certain number of people, surely something like this should not be added to repositories to avoid problems, these are just my tests ! I will try further, I have tried to pre-convert my original .mp3's to 22050 Hz , 48kb/s and even lower, but this does not affect the final result ! I will try to watch .RAW with Audacity !

[VK3KYY]

Thanks for the reply ! I realise that I am doing this as a test and experiment for a certain number of people, surely something like this should not be added to repositories to avoid problems, these are just my tests ! I will try further, I have tried to pre-convert my original .mp3's to 22050 Hz , 48kb/s and even lower, but this does not affect the final result ! I will try to watch .RAW with Audacity !

OK.

You could publish your VPR files elsewhere for people to download.

BTW.

There are some accent controls available via AmazonPolly which is the backend used by TTSMP3.

I know the French VPR texts have some accent controls in them, but I don't know if any other languages use this feature

[UB3GDZ]

It seems indeed my .RAW file is at 44100 if Audacity is giving me the information correctly. But in the script it is 8000 - ‘’+gain+‘dB’,‘-ar’,‘8000’,‘-f’,‘s16le’,outFile]’ and then it is not clear how other faster packages build and work

Снимок.JPG (78.07 KiB) Viewed 845 times

[VK3KYY]

I presume you mean, that if you open the RAW files with that setting, then Audacity plays the audio at the correct speed etc.

If this is the case, then check the conversion parameters for ffmpeg. They work OK with the MP3 files from TTSMP3, so the presumption is that ffmpeg could output the file with the same RAW paramaters , regardless of the mp3

[UB3GDZ]

I presume you mean, that if you open the RAW files with that setting, then Audacity plays the audio at the correct speed etc.

If this is the case, then check the conversion parameters for ffmpeg. They work OK with the MP3 files from TTSMP3, so the presumption is that ffmpeg could output the file with the same RAW paramaters , regardless of the mp3

If I play the .RAW file in Audacity as it is defined, namely in 44100 Hz, it plays in accelerated mode, as soon as I manually change this value to 8000Hz, it sounds fine ! Apparently I need to somehow disconnect the script from TTSMP3 and work only locally !