feat(streaming): add universal-3-6 speech model

[zkleb-aai]

Adds universal-3-6 to the streaming SpeechModel enum, alongside the existing universal-3-6-pro. This is the non-pro tier of the same generation; the enum previously had no way to select it, so callers had to bypass the type and pass a raw string.

Mirrors #233, which added universal-3-6-pro, and matches assemblyai-node-sdk#182, which shipped the same model in 4.41.1.

Changes

• assemblyai/streaming/v3/models.py — add universal_3_6 = "universal-3-6" to SpeechModel
• tests/unit/test_streaming.py — assert the model reaches the connection URL as speech_model=universal-3-6
• Version bump 1.5.2 → 1.5.3

Usage

from assemblyai.streaming.v3 import SpeechModel, StreamingParameters

params = StreamingParameters(sample_rate=16000, speech_model=SpeechModel.universal_3_6)

Testing

• pytest tests/unit — 651 passed
• ruff check / ruff format --check clean
• Verified against the live streaming service: connecting with speech_model=universal-3-6 returns a Begin frame and streams Turn events with accurate transcripts, while an invalid model name returns an Error frame — so the service recognizes this model rather than silently ignoring the parameter

🤖 Generated with Claude Code