1173667 - Sound quality of Text to speech function is poor

Status: RESOLVED WONTFIX

(Firefox OS Graveyard :: Gaia::System::Accessibility, defect)

Description

[Viga]

Using 
var u = new SpeechSynthesisUtterance();
     u.text = 'Hello World ';
     u.lang = 'en-US';
     u.rate = 1.0;
     u.onend = function(event) { alert('Finished in ' + event.elapsedTime + ' seconds.'); }
     speechSynthesis.speak(u);
to do a Text to Speech.  
The quality is not good, robotic and choppy. 
Compared to Android TTS sound quality, FFOS needs to do better.

Comment 1

[Wesly Huang (EPM)]

Hi Viga:

Would you also put some detail about the environment under test? Like the device/OS used? and which text language were you using during the test?

Comment 2

[Viga]

I'm sure that it's a generic issue. Not any Single words or sentences, not devices dependent.
And I've already mentioned it. the words I used:
 u.text = 'Hello World';
Please, give us any useful feedback ASAP.

Comment 3

[Wesly Huang (EPM)]

Viga, after double check I understand current TTS voice is not at the same quality level as other OS's, and as of today we don't yet have plan to add any new higher quality TTS voices. 

I assume you have an expectation in mind on a targeting HW spec., plus a timeline for it. Would you share them so we can better understand the requirement then know the feasibility to reach it? For example we may start w/:

1. What examples of existing text-to-speech engines do you consider "good enough"?
2. How much memory does your targeting device have? How much long-term storage space does the device have?
3. Is the device you are targeting always connected to the internet? What's the connection bandwidth? Does the data transferred cost the user per GB or is it a flat-rate?
4. What language(s) have to be supported? What language(s) have to be supported in future?

BTW it's mentioned this bug is for some feature strategy decision, when will the decision be made?

Comment 4

[Viga]

The target quality is at least Android OS's.
That's all I'm cared about based on current implementation.

Comment 5

[Wesly Huang (EPM)]

Hi Viga:

There is no easy answer for this w/o more background. I got a call w/ Fei for some context and will check w/ team then back for update. Thanks.

Comment 6

[Wesly Huang (EPM)]

Hi Viga:

We see the main limiting factor of TTS voice quality is the algorithm, in other words the TTS library used. 

It's pico TTS that FxOS currently using, and the company who built it is already acquired by others some years ago, hence as a short-term we don't see the feasibility to improve it. Yet we will consider to survey new alternatives in the future. (questions in comment#3 were meant for getting inputs so we take them into consideration, too)



The majority of my questions were geared towards deciding when (what is their time-frame) and how (memory, connectivity, target device....)
to upgrade our algorithm. Also my questions were geared towards probing if the time-frame and other details of our upgrade can be made to
coincide with TCL's requirements.

From your other comments, I gather they want a new TTS engine today. If this is the case, they need to consider TTS outsourcing, as we have
no current plans to upgrade the TTS engine, though we will do so in the near future.



Android in the beginning (<= 2.2) used picotts as well, but they built their own voice for it (side-by-side to the synthesis algorithm, a "voice" is required, that is nothing more than hours of speech recorded by only one "professional' speaker).

Comment 7

[Wesly Huang (EPM)]

(sorry, please ignore previous comment)

Hi Viga:

We see the main limiting factor of TTS voice quality is the algorithm, in other words the TTS library used. 

It's pico TTS that FxOS currently using, and the company who built it is already acquired by others some years ago, hence as a short-term we don't see the feasibility to improve it. Yet we will consider to survey new alternatives in the future. (questions in comment#3 were meant for getting inputs so we take them into consideration, too)

Comment 8

[Wesly Huang (EPM)]

Closed for now per comment#7.