Repository navigation
Add SpeechClarity, TurnTaking, and VoiceTaskSuccess voice scorers - #229
Closed
kev (kevchoi) wants to merge 2 commits into
Closed
kev (kevchoi) wants to merge 2 commits into
kev (kevchoi) wants to merge 2 commits into
Conversation
LLM-as-a-judge scorers that listen to a voice call recording instead of a transcript. Any LLM classifier now accepts an OpenAI input_audio `audio` arg, sent as a user message after the prompt. The voice templates default to gpt-audio. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
SpeechClarity and TurnTaking use the agreed prompts, and their templates declare requires_audio so they fail fast without a recording. VoiceTaskSuccess judges the call from the trace's thread_with_system. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Braintrust eval report
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds three voice scorers. Two listen to the call recording itself, not a transcript:
SpeechClarity(audio): how easily a phone listener can understand every word the agent says.TurnTaking(audio): whether the agent talks over the caller, cuts them off, or ignores interruptions. Response latency is ignored.VoiceTaskSuccess(transcript): whether the agent correctly completed the caller's request, judged from{{thread_with_system}}(instructions, tool calls, tool results).Grades are A = 1, B = 0.5, C = 0.
How
audioargument in OpenAI'sinput_audioshape ({data, format}). It's sent as a user message after the prompt, in both TS and Python.requires_audio: true. Those scorers fail fast withoutaudioinstead of judging nothing.gpt-audio. The gpt-5 models can't take audio input. OpenAI audio input accepts only WAV and MP3.Testing
SpeechClaritysends audio togpt-audio; audio is required) and Python (TurnTakingrequest shape,VoiceTaskSuccessrenders the system prompt, audio is required).mainwithout a key (25 Python, 8 JS: missing credentials orlitellm).Draft. The product wiring is in the companion braintrust monorepo PR.
🤖 Generated with Claude Code