closes #30 /claim #30

Summary

This PR implements the complete golem:stt@1.0.0 WIT interface across 5 major speech-to-text providers, delivering enterprise-grade transcription capabilities with unified APIs, real-time streaming, and graceful degradation for production use.

All providers implemented All 4 Environment Variables implemented ( STT_PROVIDER_ENDPOINT, STT_PROVIDER_TIMEOUT, STT_PROVIDER_MAX_RETRIES, STT_PROVIDER_LOG_LEVEL)

Provider-Specific

Google

GOOGLE_API_KEY=your-api-key GOOGLE_CLOUD_PROJECT=project-id

Azure

AZURE_SPEECH_KEY=your-speech-key AZURE_SPEECH_REGION=your-region

AWS

AWS_ACCESS_KEY_ID=your-access-key AWS_SECRET_ACCESS_KEY=your-secret-key AWS_REGION=your-region AWS_S3_BUCKET=transcription-bucket

Deepgram

DEEPGRAM_API_KEY=your-api-key

Whisper

OPENAI_API_KEY=your-openai-key

All providers tested against:

• ✅ Basic transcription (WAV, MP3, FLAC formats)
• ✅ Word-level timing and confidence scores
• ✅ Speaker diarization (where supported)
• ✅ Streaming transcription (4/5 providers)
• ✅ Error mappings (invalid inputs, rate limits, network errors)
• ✅ Edge cases (silence, overlapping speakers, long audio)
• ✅ Golem durability API integration

Test Commands

Build all components

cargo make build

Test individual providers

golem worker new test:stt/worker-google –env GOOGLE_API_KEY=xxx golem worker invoke test:stt/worker-google test1 –stream

Implementation Status

⚠️ = Gracefully degraded (returns appropriate none/fallback values)

Architecture

Component Structure

stt/ ├── stt/ # Core WIT interface types ├── google/ # Google Cloud Speech impl ├── azure/ # Microsoft Azure Speech impl ├── aws/ # Amazon Transcribe impl ├── deepgram/ # Deepgram API impl ├── whisper/ # OpenAI Whisper impl └── wit/ # WIT interface definitions

Generated Components

• golem-stt-google.wasm (+ portable version)
• golem-stt-azure.wasm (+ portable version)
• golem-stt-aws.wasm (+ portable version)
• golem-stt-deepgram.wasm (+ portable version)
• golem-stt-whisper.wasm (+ portable version)

Error Handling & Resilience

Comprehensive error mapping:

• invalid-audio - Malformed audio data
• unsupported-format - Unsupported audio formats
• unauthorized - Invalid API credentials
• rate-limited - API quota exceeded with retry timing
• service-unavailable - Provider downtime with retry logic
• network-error - Connection failures with backoff

Retry Strategy:

• Exponential backoff with jitter
• Configurable max retries (default: 3)
• Smart retry on transient errors only

Performance & Durability

• Real-time streaming with sub-second latency for supported providers
• Golem durability integration for crash recovery and state persistence
• Memory efficient audio chunk processing
• Concurrent request handling with connection pooling

Quick Start

Build components

cd stt && cargo make build

Deploy test application

cd test/stt && golem app deploy -b google-debug

Create worker and test

golem worker new test:stt/worker –env GOOGLE_API_KEY=xxx golem worker invoke test:stt/worker test1 –stream

Compliance

✅ Full bounty compliance:

• Implements complete golem:stt@1.0.0 WIT interface
• All 5 target providers implemented
• WASI 0.23 compatible components
• Environment variable configuration
• Golem durability API integration
• Comprehensive test coverage
• Graceful feature degradation

Videos

Google

https://github.com/user-attachments/assets/15fc2dae-fdf2-496b-a852-5a29c9ca59d4

AWS

https://github.com/user-attachments/assets/7a4072a5-d839-4cfe-816b-162a0e01ea8b

Azure

https://github.com/user-attachments/assets/45baab10-6be9-489e-b422-dc50da795fb4

Deepgram

https://github.com/user-attachments/assets/932e4a20-b795-4c99-a62f-0959299b901b

Whisper

https://github.com/user-attachments/assets/5f5d5040-9565-43d4-911f-ad2d8d1f1dc4