Issue with live transcription

[uttam-scai]

When there is no audio, the client() keeps on re-transcribing past audio.
`from whisper_live.client import TranscriptionClient
client = TranscriptionClient(
"localhost",
9090,
lang="ja",
translate=False,
model="small",
use_vad=False,
)

result = client()
print(result)`

#result: に厳しい処分が必要と答えた人が 61%に登ることが世論調査で分かりました調査はこの土日にご覧の方法で行いました岸田内閣を 指示できるという人は、前回の調査から0.1ポイント下落し、 22.8%で、6ヶ月連続で過去最低を更新しました。

The audio stops after 20 seconds but the results keep getting modified with past texts. How to solve this issue?
thank you !!

[makaveli10]

@uttam-scai I see that you use WhisperLive with a microphone. So, we consider the last segment recieved from the model as something which might change because in the current audio chunk there is a high possibility that the last word might have been cut and not fully captured in the current audio chunk so, i hope you only see the last part of the transcription which changes and not all of it.

So, if you keep talking on the microphone the transcription must stabalise and only part of it should be prone to change.

[uttam-scai]

@makaveli10 thank you for your quick response. Actually when the audio stops completely for sometime, the print result fluctuates with repeating previous words.
Is it possible to keep the entire transcript ? May you please guide me to any code that can help me to save entire conversation transcript till the mic was on?
Thank you !!
I want to pass the transcript to an LLM for my work. It will be a great help #thankyou

[makaveli10]

#324