Kindle TTS: Would you use a different voice?

[Kindle Addict]

I've recently been poking around the Kindle's filesystem and came across the following directories:

/usr/lib/tts/speech/rssolov4/
/usr/lib/tts/speech/components/

The voice files contained within these directories are almost identical to voice files available as part of the Nuance RealSpeak demo and additional voices you can purchase here.

The only difference is that the purchased files come with a Windows .dll, while the Kindle requires a Unix .so (e.g. /usr/lib/tts/speech/components/voicefont/vf_tom_full.so). I'm still looking for a source for alternate voices, but if there were, would anyone use them? I would create an installer that would replace the default voice on the Kindle with a different one.

All of this TTS investigation led me to start a write-up on TTS for the Kindle; in Part 1, I interview Tom Glynn - the male voice of the Kindle (yes, I know Engadget interviewed him a year ago, but I had different questions).

I've also attached the interview file, so you can play it pack in Tom's "own" voice, which is kind of funny (to me at least).

Thanks,
Kindle Addict

[poshm]

I do like the TTS feature on the Kindle and often use it when I'm in the middle of an exciting read but I have to go and prepare a meal or get on with housework. That way I can still continue with the story.

The male voice (Tom) is pretty good - much better than I was expecting.

But I wish I could have a voice in a British accent - just because it would be more comfortable to listen to. So yes there is a need (maybe a niche one) for other voices.

[Gwen Morse]

I'd love a female voice. The free TTS voice on my android phone is pretty good when handling turn-by-turn directions. I'd love something with a similar sound to read my ebooks.

[Kindle Addict]

Well, here's a complete list of voices from Nuance's demo site:

It seems that there are many Kindle users who would like to try out different voices. Here's a complete list of voices at Nuance (available to try via their interactive demo):

American English (Tom, Donna, Jill, Samantha)
Australian English (Karen, Lee)
Basque (Arantxa)
Belgian Dutch (Ellen)
Brazilian portuguese (Raquel)
British English (Daniel, Emily, Serena)
Canadian French (Felix, Julie)
Catalan (Nuria)
Czech (Zuzana)
Danish (Ida)
Dutch (Claire)
Finnish (Mikko)
French (Sebastien, Virginie)
German (Beate, Steffi, Yannick)
Greek (Alexandros)
Hong Kong Cantonese B5 (Sin-Ji)
Indian English (Sangeeta)
Irish English (Moira)
Italian (Paolo, Silvia)
Japanese (Kyoko)
Korean (Narae)
Mandarin Chinese B5 (Mei-Ling)
Mandarin Taiwan (Ya-Ling)
Mexican Spanish (Javier, Paulina)
Norwegian (Stine)
Polish (Agata)
Portuguese (Joana, Madalena)
Russian (Katerina, Milena)
Scottish English (Fiona)
South African English (Tessa)
Spanish (Diego, Isabel, Monica)
Swedish (Alva)
Turkish (Aylin)

If anyone has a RealSpeak ARM-based voice from another product (GPS, Windows Mobile, etc.) please post here.

Thanks,
Kindicted

[HarryT]

This sounds like the same list of voices that my Garmin GPS has; the names all look very familiar!

[Kindle Addict]

Originally Posted by HarryT

This sounds like the same list of voices that my Garmin GPS has; the names all look very familiar!

If it's a nüvi series, then it may have an ARM processor. If the voices are indeed the same, it may be possible to use the Garmin .so binaries to drive speech on the Kindle. The following article offers some hope:

"...[the nüviphone] is profoundly easy to root and it is capable, to a great extent, of running precompiled debian arm binaries."

Thanks Harry!

[HarryT]

Yes, it's a Garmin nuvi GPS. The voice files can all be downloaded from Garmin's web site, if that's of any interest.

[loveangel]

Yes, it's a Garmin nuvi GPS. The voice files can all be downloaded from Garmin's web site, if that's of any interest.

I have downloded these voices frm garmin websites

[fugazied]

Originally Posted by Gwen Morse

I'd love a female voice. The free TTS voice on my android phone is pretty good when handling turn-by-turn directions. I'd love something with a similar sound to read my ebooks.

Female would be great. I know on my mac the female voices are less grating on the ears.

[Kindle Addict]

HarryT and loveangel

I may be a bit dense here; the only voices I can find on the Garmin site that match the RealSpeak voices are for a StreetPilot 2730. The voices I can find for the Nuvi are obviously not the ones I want, since they have names like "Squirrely".

On a whim, I downloaded the Daniel voice for the 2730 and extracted the files. Besides being for a Windows-based device, the sampling rate is 16Khz, while the Kindle voices are 22KHz. This wouldn't normally be a problem, except that the Kindle seems to only support 22KHz voices. On the plus side, the Tom and Daniel DAT files have identical structures, so they should be compatible.

If you could post a link to the Nuvi voices, I'd be appreciative.

[=X=]

So I did as you suggested and listed to the interview on the commute to work. I have to say it was a great interview!

[Kindle Addict]

Originally Posted by =X=

So I did as you suggested and listed to the interview on the commute to work. I have to say it was a great interview!

Glad you enjoyed it, and thanks for posting!

Here's an update to the GPS voice progress: TomTom GPS units use the same processor as the Kindle, and have TTS capabilities. I did manage to get my hands on the "Tom" voice for the TomTom, but sadly, the Nuance Vocalizer Mobile v5 used for GPSes and Phones, etc. is missing some critical files required for the Kindle's RealSpeak Solo.

I've also analyzed the Windows versions of the US Tom and UK Daniel RealSpeak voices (demos available here), and there is literally only one required file that is different (vf_tom_full.dll instead of vf_tom_full.so). Unfortunately, these files are voicefont shared libraries, which, from what I can surmise, are Markov-chain implementations for phoneme weighting and selection - practically impossible to re-create without the source, and they vary per voice.

I'll update my status if I make additional progress, but any insights would be appreciated.

[Kindle Addict]

Well, some progress (sort of). I wanted to try out the Daniel (UK) vf file on the Kindle, so I SCPd it to the /usr/lib/tts/speech/components/data directory, only to find that the Kindle doesn't have enough internal memory to store more than the two voices that it comes with (each voice is about 100MB). I moved the Samantha voice to the /mnt/us partition (the flash partition that you see when you connect your Kindle), then moved the Daniel voice files in.

I tried replacing Tom with Daniel (i.e. re-naming the Daniel voice files as if they were Tom), but I received an "unexpected error" (see the gif attachment).

Then I tried restoring the Tom files one by one, starting with the us_tom_full.dat, then trying the speech between each one. Well, it worked, but the speech is garbled - but you can tell that it is Daniel's voice. Listen to the attached sound bite to hear the results.

So the us_tom_full.dat file is connected to the vf_tom_full.so shared library. No surprise, but it was interesting that RealSpeak had no issues using Daniel's 80MB dat voice file as if it were Tom's. This just underscores the fact that if I obtain a Daniel (or any other voice) .so file, TTS will be more than happy to speak using the replaced voice.

As always, comments or suggestions are welcome. Maybe this thread belongs in the developer area now?

[RealAct]

Any progress on this? I would love for my Kindle to read in Spanish.

Does anybody know of a hack for this? It would be awesome.

[NoLearningLimits]

I like the voices, but want the speech to be a lot faster. Now the Kindle has only three speed choices. Slow, normal, and a little faster. The fastest is about 200 words per minute. A normal reader reads at about 300 to 350 words per minute. How can we speed it up.