Based on your question about handling 8K audio for streaming transcription in Hindi, English, and Hinglish, here are the key points:

You do not need to upsample your 8K audio to 16K before sending it to Deepgram. When using raw audio streams, you should specify the correct sample rate (8000 Hz in your case) using the sample_rate parameter

Here's what you should do:

When streaming raw audio, make sure to set both:
• encoding parameter (like linear16)
• sample_rate=8000
• Since you're using raw audio, you'll need to explicitly tell Deepgram about your audio format. This is required for Deepgram to properly decode your stream
However:
• Using 16 kHz can help if the true source is wideband and you control capture; some empty/poor results have been fixed by matching the correct sample rate, occasionally trying 16 kHz instead of 8 kHz when the stream really is wideband (RTP empty transcripts tip).
• For low-quality telephony audio, consider one of our phonecall models for better robustness; it’s trained for challenging call audio (Low level audio guidance).

This message was sent by John Vajda from Deepgram, via our community automation.