A live transcription connection kept dying with received 1011 (internal error) Deepgram did not receive audio data or a text message within the timeout window. See https://dpgr.am/net0001. Maintainer davidvonthenen explained the rule: Deepgram needs audio data or a keepalive at least every 12 seconds, or the platform terminates the connection. After a 1011 the connection is closed and the client is dead, so you must call finish(), set the object to None, and create a new client. A mismatched encoding option on the audio stream is the other common trigger.

The 1011 error means Deepgram went about 12 seconds without receiving audio data or a keepalive, so the platform closed the connection. Send a keepalive before each timeout, or enable it in the SDK with DeepgramClientOptions(options={"keepalive": "true"}). Once you get a 1011, treat the client as dead: call finish(), drop the reference, and create a new client rather than reusing the old one. Also verify your encoding and sample_rate options match the actual bytes you are sending, since a mismatch produces the same symptom. Source: https://github.com/deepgram/deepgram-python-sdk/issues/319