speechall

CLI for speech-to-text transcription via the Speechall API.

Supports providers like OpenAI, Deepgram, AssemblyAI, Cloudflare, Groq, ElevenLabs, Google, Gemini, and more.

Install

Homebrew (macOS and Linux)

brew install Speechall/tap/speechall

From source

Requires Swift 6.2+.

git clone https://github.com/Speechall/speechall-cli.git
cd speechall-cli
swift build -c release
cp .build/release/speechall /usr/local/bin/

Setup

Get an API key from speechall.com and either export it or pass it directly:

export SPEECHALL_API_KEY="your-key"

Usage

# Transcribe an audio file
speechall audio.wav

# Specify a model
speechall audio.wav --model deepgram.nova-2

# Choose output format
speechall audio.wav --output-format srt

# Enable speaker diarization
speechall audio.wav --diarization --speakers-expected 3

# Pass API key directly
speechall audio.wav --api-key your-key

# List available models
speechall models

# Filter models by provider or capability
speechall models --provider deepgram
speechall models --language tr
speechall models --diarization --srt

Querying models with jq

speechall models outputs JSON, so you can pipe it to jq for advanced queries:

# Cheapest Deepgram model with SRT support
speechall models --provider deepgram --srt | jq 'sort_by(.cost_per_second_usd) | .[0].id'

# All model IDs that support Turkish
speechall models --language tr | jq '[.[].id]'

# Compare diarization models by price
speechall models --diarization | jq '[.[] | {id, cost: .cost_per_second_usd}] | sort_by(.cost)'

# List all providers
speechall models | jq '[.[].provider] | unique'

# Models that support both streaming and custom vocabulary
speechall models --streamable --vocabulary | jq '[.[].id]'

Run speechall --help for all options.

License

MIT