Actually, FTTH is back.
So two months after layoff and while we were unable to know if Mozilla would continue DeepSpeech
Link please? Josh worked on this feature. I’m not sure what discussion you are referring to.
What issue ?
For smaller domain specific language models i found that boost values in the range of 1-20 gave me sufficient results to improve the recognition of the wanted words greatly. Mostly I stay in the 1-10 range actually.
Anything higher than that gave me worse results.
Keep in mind that a boost word doesn’t work with a space in it.
But thats just my personal experience thats limited to smallish language models.
Do you really need the general language model though? It really sounds to me like you have a very specific use case?
Wouldn’t it be easier to create a scorer from just the sentences that you actually expect?
If you are looking for off-the-shelf plug-and-play 24-7 support, sorry, but this is not what we can yet provide.
We have a general purpose 3D scene application, and we use standard language to control it, but then certain content also has custom words, crater names, biological words, etc
Come on, you don’t even know the parameter usage for your own functions! This is just poor management, and why rockets and planes blow up and crash, and maybe why Mozilla bosses gave up, I wonder. Let’s not make excuses, you need to sort this critical function out.
Your posts about the boost value for addHotWords is titled “enable hot-word boosting #3297”. It’s on your github.
This is a long PR, I’m not sure I remember all of it.
Rudeness is not going to get you any help, you know. Can we please focus on what is actionable?
1 Like
Try testing that with words like schrödinger (or schroedinger), schumacher and tycho
It does looks like you still don’t understand: we are not working anymore on this as part of our job.
I currently don’t have much spare time to allocate to work on this kind of feature, sorry. I burnt myself over the 2020 year, especially preparing the 1.0 release, that layoff wrecked, so now I can’t push too much and I physically need resting.
Software engineering is hard and needs tough people at times to see it through. It’s not always a “fun job” as is being pushed by the media on kids.
If you’re burned out, take a break, but it would be a terrible waste to see this project fail because there isn’t one tough-minded manager around to take it forward, so don’t give up.
Thanks, I’m young and weak then.
1 Like
So run the generate lm on the non augmented corpus and see what the least common word in the top 50k is and than add your words with more repetitions than that to the corpus.
I think your looking for an easy off the shelf solution and commercial support and you just wont get this here on an open source project that has no official resources allocated to it anymore.
I for one am really thankful for the few people like @lissyx who keep this project above water against the odds in their own free time in addition to their day job tasks.
To give you perspective learning how to build a custom scorer for deepspeech and putting in the effort to understand the concepts really is a walk in the park compared to doing the same for other opensource stt tools like kaldi.
If your not willing to put in that effort to learn how to use the great tool that deepspeech is you should really look at commercial solutions as in the opensource stt world it just gets more complicated than deepspeech.
1 Like
I’m not saying that, maybe though you haven’t had enough support, but clearly things are a bit of a mess at the moment, but a few days hard work could put it right, if any of the core engineers are still about.
JFKK: Well let’s be realistic, because that’s what keeps you on the payroll, who wants to use a complex speech tool where the project team don’t know how their own functions work?!
What do you mean with payroll? I have no affiliation with deepspeech or mozilla whatsoever.
Im not even a software engineer in my day job.
I use deepspeech because open source stt is my hobby and i work with a non profit in my freetime that builds voice automated homes for severely handicapped people here in germany.
I simply use deepspeech because i like the apis, performance, ease of use for adaption and the project in general. I was using kaldi before that and before that pocketsphinx.
1 Like
I meant we need to be realistic, look at all the layoffs at Mozilla! Doing the job properly keeps people employed, having a Speech API where nobody in the team knows the parameters is just ridiculous. Some 60% of software projects fail due to a lack of adequate management. The addHotWord function is critical to many applications, it needs to be fixed in a reasonable time, and 5 months has gone by with no action. Even on an open source project there has to be some order and discipline or we’re just wasting everyone’s time.
This is plain wrong …
It is so critical, that nobody before Josh PR took the time to work on such a feature, so I guess it means not so many people actually need it
What action do you expect? Please articulate the problem.
Like, when requesting support, giving a proper context so that people who are helping knows what you do? Like the fact you are working on Windows and we document we don’t support training there?
Thats actually one thing i can agree on. The hotword feature is very very useful especially when working with weaker models like the german one. Its been a game changer in my application for deepspeech.
3 Likes
I’ve never said it can’t be a game changer to some use-case, but given all the people who shared feedback / sent feature requests, etc., we have received bigger PRs than this one much much earlier.
Reminder: the team, before layoffs, was already small (5 persons), so it’s not like we can solve everything at once.
1 Like