S15 · Pipeline skill
Connect speech services
Turn speech into reviewable text, respond with a selected voice and inspect the session.
Speech transcription + text understanding + human checks and speech output
- Pipeline
- 4 steps
- Human review
- 1
- Components
- 4
01Pipeline
Base pipeline
Ordered steps. Each consumes the previous output; review gates hold until a person accepts.
- Starts from
- Consented audio, language and output mode
- Delivers
- Session events and usage records
Review contract
People decide at review gates; components supply data, drafts or review surfaces. Unaccepted outputs do not advance.
Integration note
Transcription, text analysis and realtime transport are separate interfaces; orchestrate buffered or realtime paths explicitly.
- 01
Transcribe and interpret
- Input
- Consented audio, language and output mode
- Output
- Transcript, topic and intent candidates
- Speech transcription and audio translation adaptersAudio transcription or translation + language and format controls
- Provider text summarization, sentiment and intent analysisSupplied text topic and intent extraction + structured summaries
- 02
Check reply content
Human review- Input
- Transcript, topic and intent candidates
- Output
- Human-confirmed reply text
Decided byAssigned reviewer or requesting user
- Provider text summarization, sentiment and intent analysisSupplied text topic and intent extraction + structured summaries
- 03
Generate spoken reply
- Input
- Human-confirmed reply text
- Output
- Reply audio and format information
- Configurable speech synthesis relayText-to-speech requests + output format adaptation
- 04
Check session handoff
- Input
- Reply audio and format information
- Output
- Session events and usage records
- Realtime audio proxy and provider session integrationBidirectional audio event proxy + session state and usage
02Industry versions
Industry versions
Healthcare Transferable use Add transcription and speech to appointment inquiries, with staff checking important dates before reading them back.
Adaptation Administrative questions only; transcription, the human confirmation interface and realtime sessions need explicit orchestration.
- 01
Transcribe appointment needs
- Input
- Consented appointment inquiry audio and language settings
- Output
- Appointment text and intent candidates
- Speech transcription and audio translation adaptersAudio transcription or translation + language and format controls
- Provider text summarization, sentiment and intent analysisSupplied text topic and intent extraction + structured summaries
- 02
Staff check the reply
Human review- Input
- Appointment text and intent candidates
- Output
- Confirmed administrative service reply
Decided byAssigned reviewer or requesting user
- Provider text summarization, sentiment and intent analysisSupplied text topic and intent extraction + structured summaries
- 03
Read the checked reply
- Input
- Confirmed administrative service reply
- Output
- Appointment reply audio
- Configurable speech synthesis relayText-to-speech requests + output format adaptation
- 04
Check voice session records
- Input
- Appointment reply audio
- Output
- Session handoff and usage records
- Realtime audio proxy and provider session integrationBidirectional audio event proxy + session state and usage
03Parameters
Parameters
From the pack’s params.schema.json. Choose values to build a synthetic request.
skills/_pipelines/connect-speech-services/params.schema.json
{
"synthetic": true,
"run_id": "synthetic-example",
"locale": "en",
"industry": "base",
"adapter_mode": "local",
"records": [
{
"id": "synthetic-record-1",
"synthetic": true,
"text": "Synthetic observation A: a draft needs review."
},
{
"id": "synthetic-record-2",
"synthetic": true,
"text": "Synthetic observation B: retain the unresolved question."
}
],
"decisions": {}
} 04Member components
Member components
-
Speech transcription and audio translation adapters
Audio transcription or translation + language and format controls
Working 1 pipeline -
Provider text summarization, sentiment and intent analysis
Supplied text topic and intent extraction + structured summaries
Working 2 pipelines -
Configurable speech synthesis relay
Text-to-speech requests + output format adaptation
Working 1 pipeline -
Realtime audio proxy and provider session integration
Bidirectional audio event proxy + session state and usage
Prototype 1 pipeline
05Skills pack
Skills pack
Runnable reference on synthetic fixtures with tests. Domain contracts, services and approvals still need integration.
connect-speech-services-1.0.0.zip
Contains SKILL.md, pipeline.json, params.schema.json, runner, tests, fixtures, integration checklist and sources.
Synthetic workflow reference; service integration pending
Download · 17KBSHA-256 9db03e6dea239928c01518c5c1401ff5e294edc3b016f8d5ab4f7b7249ff6b7e