pypi / assemblyai
assemblyai
What changed in assemblyai 0.64 to 1, from its official release notes, upgrade guides and advisories. Sources checked 2026-11-12.
7 breaking33 new APIs and options4 behavior and defaults1 tooling8 fixes and other
See which of these apply to your code, ranked, in your agent:
npx @unstale/cli initassemblyai 143 changes
Breaking changes 6
The live connection becomes the only connection the sync client opens.
The LeMUR API and all aai.Lemur* names are removed; send transcript text to an LLM or use aai.LLMGateway instead.
The extras install option is removed; installing with this package extra fails.
The extras module, including MicrophoneStream, is removed; capture audio with another library and pass PCM to the streaming client.
LeMUR support was removed from the SDK.
RealTimeTranscriber was introduced as part of a transcriber rename.
New APIs and options 27
The streaming API accepts the veterinary-v1 domain.
Adds a Text-to-Speech HTTP client for developers to use.
Adds a WebSocket session client for developers to use.
DictationConfig gains rewrite_mode, allowing developers to configure dictation rewrite behavior.
Adds an LLM Gateway module.
Adds the speaker_labels_revision_interval_ms parameter to streaming.
Adds the acknowledge_silence parameter for streaming.
Adds the Silence event for streaming.
Adds TranscriptConfig to improve the Speech Understanding UX.
Adds the universal-3-6 speech model for streaming.
Adds the Dictation API, with live upload as its only connection.
Adds stt_prompt to DictationConfig.
Adds push-style live sessions via open_live().
Adds transcribe_stream() to stream the sync upload.
Adds a localization option to language_detection_options.
Adds short_file_diarization_method to SpeakerOptions.
Adds the universal-3-6-pro speech model for use with the SDK.
Client, transcriber, and streaming constructors now accept api_key=.
A versioned subpackage provides the preferred import path for prerecorded transcription APIs.
A versioned subpackage provides the preferred import path for streaming APIs.
A versioned subpackage provides the preferred import path for sync transcription APIs.
Added AsyncSyncTranscriber for asyncio callers.
Async transcribers can be used as async context managers to release their HTTP connection pools; aclose() is the explicit equivalent.
The streaming surface gains RealTime* names as the preferred names for the former Streaming* classes; old names remain bound to the same objects.
SyncTranscriber gains warm() to establish a connection before transcription; AsyncSyncTranscriber has an awaitable equivalent.
The transcribe methods accept poll_timeout to bound polling and prevent a stuck job from hanging a request handler.
Transcribers accept a per-client API key, avoiding the need to configure process-wide state; when combined with client=, the key takes precedence.
Behavior and default changes 3
TTS sessions share one TLS context and HTTP connections are kept open for 120 seconds.
Raises the sync and dictation configuration caps to match the service.
Local files can now be passed as pathlib.Path objects directly, without converting them to strings.
Tooling 1
Ships a PEP 561 py.typed marker so type checkers can recognize the package's type information.
Fixes and other changes 6
Calling cancel() ends a reader after the request is written.
Typing was added or fixed for LLMGW Speech Understanding Summarization requests through the TranscriptConfig API.
Non-negative validation constraints were added to a RealTimeTr… component.
TLS reads are held until the handshake request is written, fixing sync client streaming behavior.
Guards model_config so the pydantic v1 fallback keeps its config.
Fixes the scan so it no longer drops later channels.
assemblyai 0.652 changes
Breaking changes 1
Prerecorded transcription was moved into prerecorded/v2, changing where developers access it.
New APIs and options 1
Adds AsyncTranscriber for callers using asyncio.
assemblyai 0.648 changes
New APIs and options 5
MicrophoneStream adds pause() and resume() methods.
Sentences and paragraphs responses now include speech_model_used.
The transcribe type hints now accept bytes.
Opus is now supported for streaming encodings.
Synchronous connection pre-warming is now available.
Behavior and default changes 1
language_codes can now be updated mid-stream.
Fixes and other changes 2
AutohighlightResult.rank is now Optional to tolerate a missing field.
The upload type hints now accept bytes.