]> VCONIC

ghostofbasho@gmail.com

Applications and Real-Time vCon Internet-Draft This document defines the World Transcription Format (WTF) extension for Virtualized Conversations (vCon). The WTF extension provides a standardized method for representing speech-to-text transcription data from multiple providers within vCon containers. This extension enables consistent transcription storage, analysis, and interoperability across different transcription services while preserving provider-specific features through extensible fields. The WTF extension is designed as a Compatible Extension that introduces transcription attachments without altering existing vCon semantics, ensuring backward compatibility with existing vCon implementations. About This Document The latest revision of this draft can be found at . Status information for this document may be found at . Discussion of this document takes place on the vCon Working Group mailing list (), which is archived at . Subscribe at . Source for this draft and an issue tracker can be found at .

Requirements Language The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 RFC2119 when, and only when, they appear in all capitals, as shown here.

Introduction Virtualized Conversations (vCon) provide a standardized container format for conversation data, enabling interoperability across different communication platforms and modalities. As speech-to-text technology becomes increasingly important for conversation analysis, compliance, and accessibility, there is a growing need for standardized transcription representation within vCon containers. Current transcription services each use proprietary formats, making it difficult to:

• Switch between transcription providers without data conversion
• Perform comparative analysis across different providers
• Maintain consistent data processing pipelines
• Preserve transcription metadata during system migrations
• Ensure long-term accessibility and archival compliance

This document defines the World Transcription Format (WTF) extension for vCon, which addresses these challenges by providing:

• A unified schema for transcription data from any speech-to-text provider
• Hierarchical organization from words to segments to complete transcripts
• Extensible fields for provider-specific features
• Built-in quality metrics and confidence scoring
• Support for real-time and batch transcription workflows
• Consistent export capabilities to standard subtitle and caption formats

The WTF extension enables organizations to standardize their transcription workflows while maintaining the flexibility to use multiple providers and preserve provider-specific enhancements.

Conventions and Definitions

Core Terms World Transcription Format (WTF): A standardized JSON schema for representing speech-to-text transcription data from any provider in a consistent, interoperable format. Transcription Provider: A service or system that converts audio or video content to text, such as Whisper(TM), Deepgram(TM), AssemblyAI(TM), Google Cloud Speech-to-Text(TM), Amazon Transcribe(TM), or Azure Speech Services(TM). WTF Attachment: A vCon attachment with type "wtf_transcription" that contains structured transcription information in World Transcription Format. Segment: A logical chunk of transcribed content, typically representing sentence or phrase boundaries with associated timing information. Speaker Diarization: The process of partitioning an audio stream into homogeneous segments according to the speaker identity, enabling "who spoke when" analysis. Compatible Extension: A vCon extension that introduces additional data without altering the meaning or structure of existing elements, as defined in .

World Transcription Format Overview The World Transcription Format uses a hierarchical JSON structure designed to capture transcription data at multiple levels of granularity while maintaining consistency across different providers.

Design Principles Completeness: WTF captures all essential transcription data including text, timing, confidence scores, and speaker information. Consistency: Provides uniform structure regardless of the underlying transcription provider. Extensibility: Supports provider-specific features through structured extension fields. Validation: Includes built-in data integrity checking and quality metrics. Hierarchical Organization: Natural progression from words to segments to complete transcripts.

Core Structure WTF uses a hierarchical JSON structure with three required sections and multiple optional enrichment layers:

vCon WTF Extension Definition

Extension Classification The WTF extension is a Compatible Extension as defined in Section 2.5 of . This extension:

• Introduces transcription data without altering existing vCon semantics
• Can be safely ignored by implementations that don't support transcription processing
• Does not require listing in the must_support parameter
• Maintains backward compatibility with existing vCon implementations

Extension Registration This document defines the "wtf_transcription" extension token for registration in the vCon Extensions Names Registry:

• Extension Name: wtf_transcription
• Extension Description: World Transcription Format for standardized speech-to-text representation with multi-provider support
• Change Controller: IESG
• Specification Document: This document

Extension Usage vCon instances that include WTF transcription attachments SHOULD include "wtf_transcription" in the extensions array:

WTF Attachment Structure

Attachment Container Transcription information MUST be included as vCon attachments using the standard attachment object structure defined in Section 4.4 of . The WTF transcription attachment MUST include:

• type: MUST be set to "wtf_transcription"
• encoding: MUST be set to "json" for structured transcription data
• body: MUST contain the WTF transcription data structure as defined below

The WTF transcription attachment SHOULD include:

• start: ISO 8601 timestamp when transcription was created
• party: Index of the party in the vCon parties array (for single-speaker transcription)
• dialog: Index of the associated dialog in the vCon dialog array

WTF Body Structure The body field of the WTF transcription attachment MUST contain a JSON object with the following structure:

Required Fields

Transcript Object The transcript object provides the high-level summary of the entire transcription:

• text: The complete, concatenated transcription text
• language: MUST use BCP-47 format (examples: "en-US", "es-MX", "fr-CA")
• duration: Floating-point seconds, MUST be >= 0
• confidence: Normalized to [0, 1] range regardless of provider scale

Segments Array Segments represent logical chunks of transcribed content, typically sentence or phrase boundaries:

• id: MUST be unique within the document, typically sequential
• start/end: Floating-point seconds, where end > start
• text: SHOULD be trimmed of leading/trailing whitespace
• speaker: Can be integer (0, 1, 2) or string ("Speaker A")
• words: References indices in the words array

Metadata Object The metadata object captures processing and source information:

• created_at/processed_at: MUST use ISO 8601 format
• provider: Lowercase identifier for supported providers
• model: Provider's model identifier (e.g., "whisper-large-v3", "nova-2")
• options: Preserves provider-specific configuration

Optional Fields

Words Array

Speakers Object

Quality Object

Extensions Object

Provider Integration Guidelines

Supported Providers The WTF extension supports integration with major transcription providers:

• Whisper(TM): OpenAI's open-source speech recognition system
• Deepgram(TM): Real-time speech-to-text API
• AssemblyAI(TM): AI-powered transcription and audio intelligence
• Google Cloud Speech-to-Text(TM): Google's speech recognition service
• Amazon Transcribe(TM): AWS speech-to-text service
• Azure Speech Services(TM): Microsoft's speech recognition platform
• Rev.ai(TM): Automated and human transcription services
• Speechmatics(TM): Real-time and batch speech recognition
• Wav2Vec2(TM): Facebook's self-supervised speech recognition model
• Parakeet(TM): NVIDIA's speech recognition toolkit

Conversion Requirements When converting from provider-specific formats to WTF:

1. Normalize confidence scores to [0.0, 1.0] range
2. Convert timestamps to floating-point seconds
3. Standardize language codes to BCP-47 format
4. Preserve provider-specific features in extensions field
5. Validate output against WTF schema requirements

Provider-Specific Mappings

Whisper Integration

Deepgram Integration

Quality and Confidence Metrics

Confidence Score Normalization All confidence scores MUST be normalized to the [0.0, 1.0] range:

• 1.0: Highest confidence (perfect recognition)
• 0.9-1.0: High confidence
• 0.7-0.9: Medium confidence
• 0.5-0.7: Low confidence
• 0.0-0.5: Very low confidence

Provider-specific scales are converted during import:

• Percentage (0-100) -> divide by 100
• Log probability -> exponential transformation
• Custom scales -> linear normalization

Quality Metrics The quality object provides assessment metrics:

• audio_quality: Categorical assessment (high/medium/low)
• background_noise: Noise level [0.0-1.0]
• multiple_speakers: Boolean indicator of multi-speaker content
• overlapping_speech: Boolean indicator of speaker overlap
• silence_ratio: Percentage of audio that is silence
• average_confidence: Mean confidence across all words/segments
• low_confidence_words: Count of words below 0.5 confidence
• processing_warnings: Array of processing issues or notices

Security Considerations

Data Privacy Transcription data often contains sensitive personal information. Implementations SHOULD:

• Apply appropriate access controls to WTF transcription attachments
• Consider encryption requirements for transcription data at rest and in transit
• Implement data retention policies consistent with privacy regulations
• Provide mechanisms for transcription data redaction or anonymization

Provider-Specific Security When integrating with external transcription providers:

• Validate provider credentials and API security
• Implement secure communication channels (TLS 1.2 or higher)
• Consider data residency requirements for audio processing
• Audit provider data handling practices and compliance certifications

Integrity Protection WTF transcription attachments SHOULD be integrity protected using vCon signing mechanisms as defined in to prevent unauthorized modification of transcription data.

Temporal Validation Implementations SHOULD validate that transcription timestamps are consistent with the associated dialog timing information to detect potential tampering or synchronization issues.

IANA Considerations

vCon Extensions Names Registry This document requests IANA to register the following extension in the vCon Extensions Names Registry established by :

• Extension Name: wtf_transcription
• Extension Description: World Transcription Format for standardized speech-to-text representation with multi-provider support
• Change Controller: IESG
• Specification Document: This document

WTF Attachment Type Values Registry This document requests IANA to establish a new registry for WTF attachment type values with the following initial registration:

• Type Value: wtf_transcription
• Description: Structured transcription records using World Transcription Format
• Change Controller: IESG
• Specification Document: This document Registration Template: Type Value: The string value used as the attachment type identifier Description: Brief description of the attachment type and its purpose Change Controller: For Standards Track RFCs, list "IESG". For others, give the name of the responsible party. Specification Document(s): Reference to defining documents with URIs where available

WTF Provider Registry This document requests IANA to establish a new registry for WTF transcription providers with initial registrations for supported providers:

• Provider Name: whisper
• Description: OpenAI Whisper(TM) speech recognition system
• Change Controller: IESG
• Specification Document: This document (Additional provider registrations would be added for each supported provider)

Examples

Basic Two-Party Call Transcription

Multi-Provider Transcription Comparison

References Normative References In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. SIPez LLC Informative References Strolid

Acknowledgements

• Appreciation to the transcription provider community for their input on standardization requirements.
• Thanks to the vCon working group for their feedback and guidance on extension design patterns.

Trademark Notice All trademarks mentioned in this document are the property of their respective owners. The use of these trademarks does not imply endorsement by the IETF or the authors of this document. The following trademarks are referenced:

• Whisper(TM) is a trademark of OpenAI, Inc.
• Deepgram(TM) is a trademark of Deepgram, Inc.
• AssemblyAI(TM) is a trademark of AssemblyAI, Inc.
• Google Cloud Speech-to-Text(TM) is a trademark of Google LLC.
• Amazon Transcribe(TM) is a trademark of Amazon.com, Inc.
• Azure Speech Services(TM) is a trademark of Microsoft Corporation.
• Rev.ai(TM) is a trademark of Rev.com, Inc.
• Speechmatics(TM) is a trademark of Speechmatics Limited.
• Wav2Vec2(TM) is a trademark of Meta Platforms, Inc.
• Parakeet(TM) is a trademark of NVIDIA Corporation.