alphacep / alphacep/vosk-api

Python binding: strange effect when reusing KaldiRecognizer (Reset not working properly?)

オープン
#1,372 コメント 1 件 リアクション 0 件 担当者 0 名 GitHub で見る
主要言語
Jupyter Notebook
スター
15.1k
フォーク
1.8k
PR マージ指標
30日以内にマージされた PR はありません

説明

Hi,

I noticed a strange effect when I checked a set of transcriptions created by running through a number of WAV files in different folders. Sometimes I get slightly different results for the same WAV files with the exact same settings, depending on the order of processing the files 🤨.

Since I was reusing the `KaldiRecognizer` in a loop (`for wav in wav_files ...`) I thought the only logical explanation could be that there is "something" left internally that remains inside the recognizer and influences the next transcription and indeed, when I add `rec = KaldiRecognizer(model, 16000)` in **every** iteration the effect is gone and everything is always identical 🤔.
I tried to add `rec.Reset()` instead, but that didn't work, only creating a new instance every time helps.

Is this expected behavior? Shouldn't a `rec.FinalResult()` or a `rec.Reset()` always reset the recognizer properly?

コントリビューションガイド

このリポジトリのコントリビューションガイドは索引されていません

評価

この issue はまだ評価されていません。

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。