Audio buffer fix #47
Audio buffer fix #47
Conversation
Added buffer size variable Minor fixes
|
Hey! @mainvolume Thank you very much for this PR :) Could you please change the base? so your changes are merged in the development branch? I try to do it myself but it ends up adding some old commits not related to your changes. Also, it will be great if you could explain a little why this is needed? It's kind of clear looking at the code but for those not so familiar with it a short explanation will be really helpful. Could it be the case that these changes fix #44 ? Thanks again! |
|
Surething! |
|
regarding #44 Could be as the model sample-rate has to be the same device for actual decoding 😄 when streaming. Have not tested with bluetooth device, but guessing that the audio settings from the device becomes easier when not set to a static frequency and adaptable to inputbus sample rate of the device. This way, it becomes as well possible to use to the same decoder functionality with macOS as well. |
Added buffer decoding
|
Hi Bruno. I also added a decode buffer function for already obtained buffers and other streams of audio with added start and end utterance convenience functions. 🙂 |
TLSphinx/Decoder.swift
Outdated
|
|
||
| import Foundation | ||
| import AVFoundation | ||
| import Sphinx | ||
|
|
||
| public let bufferSize = 16384 |
There was a problem hiding this comment.
Where this value come from? Maybe add a comment about it?
TLSphinx/Decoder.swift
Outdated
| do { | ||
| try AVAudioSession.sharedInstance().setCategory(AVAudioSessionCategoryRecord) | ||
| try AVAudioSession.sharedInstance().setCategory(AVAudioSessionCategoryPlayAndRecord, with: [.mixWithOthers, .allowBluetoothA2DP]) |
There was a problem hiding this comment.
Maybe we should add these as parameters to startDecodingSpeech? 🤔 Maybe there isn't a fixed set of settings for the audio session that works for everybody...
There was a problem hiding this comment.
Updated to this:
public func startDecodingSpeech (_ audioSessionCategoryOptions:AVAudioSessionCategoryOptions = [.mixWithOthers, .allowBluetoothA2DP], utteranceComplete: @escaping (Hypothesis?) -> ()) throws {
do {
try AVAudioSession.sharedInstance().setCategory(AVAudioSessionCategoryPlayAndRecord, with: audioSessionCategoryOptions)
} catch let error as NSError {
print("Error setting the shared AVAudioSession: \(error)")
throw DecodeErrors.CantSetAudioSession(error)
}
| @@ -248,7 +251,33 @@ public final class Decoder { | |||
| engine.stop() | |||
| engine = nil | |||
| } | |||
|
|
|||
|
|
|||
| public func startUtterence() { | |||
There was a problem hiding this comment.
I think this and endUtterence shouldn't be public. Is my understanding that we needed public because you should call startUtterance() before startDecodingBuffer right?
There was a problem hiding this comment.
That is accurate. Shall we make the endUtterence private you mean?
Updated audio session setting in decode speech function call
| self.start_utt() | ||
| } | ||
|
|
||
| public func startDecodingBuffer(buffer: AVAudioPCMBuffer!, time: AVAudioTime!, utteranceComplete: @escaping (Hypothesis?)-> ()) throws { |
There was a problem hiding this comment.
👏🏻👏🏻👏🏻 nice!
These will be really useful. How are you testing this?
There was a problem hiding this comment.
Also, something was wrong with the tabs? jaja
There was a problem hiding this comment.
The tabs... editing in github as the codebase in home and at work right now. 😂
Havent written any tests, but to bypass the microphone usage for the thinking machine implementation, a synthesized continuous buffer is passed to the function with which works quite sweet with.
The function is based on the streaming function but with the option of creating the buffer before passing it to the function, instead of using the tap in the function.
Updated tabs
|
There, it should be tabbed cleaner now. Also, a sample project using TLSphinx (without buffer) |
|
🤔 If you wish, we can make it private, but then the utterance would be running when the buffer ends considering the start call after reading the utterance. 🙂 |
Audio buffer and format fix.