vosk | VOSK Speech Recognition Toolkit | Speech library

( First answer:
Looking at this code, I guess nr_words is a list.
Could you specify what nr_words represents ? Is it like the list of the 'already seen' words ?

I also see that you print out nr.word so I suppose that nr_words is a list of Word objects.

But, the 2nd for loop is iterating over all the values of the nr_words list (Word objects), not its indexes.
So when you compare the two Word object on line 4, I think you should simply be using nr as the other argument for your compare() method, instead of nr_words[nr].
)

EDIT:
Reply to your comment

nr_words is an kind of empty list so that when I can compare it to dictionaries and append not repeating words in nr_words. Also I tried as you said to pass nr But got error AttributeError: 'str' object has no attribute 'word'

The error is because when the two words are not the same, you append w.word, w.start and w.end to the nr_words list (which are strings, floats and floats respectively) Try by appending only the Word object like so :

Corrected Code:

Put the Model folder beside the src folder.

Signed PCM is certainly supported. The problem is that 48000 fps is not. I think the highest frame rate supported by Java directly is 44100.

As to what course of action to take, I'm not sure what to recommend. Maybe there are libraries that can be employed? It is certainly possible to do the conversions manually with the byte data directly, where you enforce the expected data formats.

I can write a bit more about the conversion process itself (assembling bytes into PCM, manipulating the PCM, creating bytes from PCM), if requested. Is the VOSK expecting 48000 fps also?

Going from stereo to mono is a matter of literally taking the sum of the left and right PCM values. It is common to add a step to ensure the range is not exceeded. (16-bit range if PCM is coded as normalized floats = -1 to 1, range if PCM is coded as shorts = -32768 to 32767.)

Following code fragment is an example of taking a single PCM value (signed float, normalized to range between -1 and 1) and generating two bytes (16-bits) in little endian order. The array buffer is of type float and holds the PCM values. The array audioBytes is of type byte.

Let us go through your code, specifically this block

If I understand the issue correctly, you probably want to replace await self.do_recognition() with asyncio.create_task(self.do_recognition()) to make do_recognition execute in the background. If you need to support Python 3.6 and earlier, you can use loop.create_task(...) or asyncio.ensure_future(...), all of which in this case do the same thing.

When doing that you'll also need to extract the value of self._buffer and pass it to do_recognition as parameter, so that it can send the buffer contents independently of the new data that arrives.

Two notes unrelated to the question:

• The code is accessing internal implementation attributes of queue, which should be avoided in production code because it can stop working at any point, even in a bugfix release of Python. Attributes that begin with _ like _finished and _unfinished_tasks are not covered by backward compatibility guarantees and can be removed, renamed, or change meaning without notice.

• You can import CancelledError from the top-level asyncio package which exposes it publicly. You don't need to refer to the internal concurrent.futures._base module, which just happens to be where the class is defined by the implementation.

It is great you tried Vosk offline speech recognition on Android, here are some answers to your questions:

actually I read it still doesn't support my native language.

If you are about Hebrew, we might support it in the future, and you can build it yourself.

is that the reason?

You didn't provide enough information to answer this, please explain a bit more what is "it is not work"

I want to know how it works?

Extensive documentation on speech recognition is available on lectures, courses and books. You can find some introduction here for example: https://www.youtube.com/watch?v=q67z7PTGRi8

does it is doing speech to text and saved it in assets files?

It does speech to text, but it doesn't save results into assets, it just displays them. You can not modify assets, they are static.

does it is run in background?

Yes

does it is used AI models?

Sure

how it behaves when two apps need mic resource in parallel?

In android it is not possible to record audio from two apps in parallel, second one will be blocked.

noises?

It is robust to noises.

does it is work with Neural Networks API?

No, it is portable

how can I start developing such a thing?

Get some basic understanding and start writing the code. If you have further questions you can ask them in the Telegram chat

I actually did it on Mac OS X.

1. Follow instructions to compile Kalid at https://alphacephei.com/vosk/install
2. Follow instruction to make C# wrapper (same)
3. Create Xcode project and make bundle.
4. Then add bundle and c# files to your unity project.

On Windows you can install vosk from this repo: