Loading…
Speech-to-Text
The agent's ears. STT must be fast, accurate, and handle real-world audio conditions — background noise, accents, cross-talk, and telephone-quality audio.
Local / Open-Source Tools
Run speech-to-text entirely on your machine — no cloud, no API keys, no audio leaving your network. These tools let you transcribe recordings, stream live audio, and benchmark model performance all locally.
whisper.cpp
C/C++ port of OpenAI Whisper. Runs on CPU/GPU locally. Best for batch processing but can do near-real-time with streaming mode.
bash
# whisper.cpp is pre-built at /opt/whisper.cpp # System-wide wrappers available at /usr/local/bin # (whisper-cli, whisper-stream, whisper-server) # Transcribe a file whisper-cli \ -m /opt/whisper.cpp/models/ggml-base.en.bin \ -f /opt/whisper.cpp/samples/jfk.wav # Stream from microphone (real-time, requires SDL2) whisper-stream \ -m /opt/whisper.cpp/models/ggml-base.en.bin \ -t 8 \ --step 500 \ --length 5000 # Compare speed between models using time time whisper-cli \ -m /opt/whisper.cpp/models/ggml-tiny.en.bin \ -f /opt/whisper.cpp/samples/jfk.wav time whisper-cli \ -m /opt/whisper.cpp/models/ggml-base.en.bin \ -f /opt/whisper.cpp/samples/jfk.wav # Color-coded confidence output (green = high, red = low) whisper-cli \ -m /opt/whisper.cpp/models/ggml-base.en.bin \ -f /opt/whisper.cpp/samples/jfk.wav \ --print-colors # Compare multilingual vs English-only on a larger model time whisper-cli \ -m /opt/whisper.cpp/models/ggml-small.bin \ -f /opt/whisper.cpp/samples/jfk.wav time whisper-cli \ -m /opt/whisper.cpp/models/ggml-small.en.bin \ -f /opt/whisper.cpp/samples/jfk.wav
Faster Whisper
CTranslate2-based Whisper reimplementation. 4x faster than original with same accuracy. Great for real-time with VAD.
Privacy Notice
The first run normally downloads models from Hugging Face. Pre-downloaded models are available at
Use
The first run normally downloads models from Hugging Face. Pre-downloaded models are available at
/opt/faster-whisper/models/Use
--model with a local path to skip Hugging Face entirely:faster-whisper audio.wav --model /opt/faster-whisper/models/base.enbash
# Faster Whisper is pre-installed and available system-wide # Wrapper at /usr/local/bin/faster-whisper → /opt/faster-whisper/transcribe.py # Managed via uv venv at /opt/faster-whisper # Basic transcription (using local model) faster-whisper audio.wav \ --model /opt/faster-whisper/models/base.en # Specify device and language faster-whisper audio.wav \ --model /opt/faster-whisper/models/base.en \ --device cpu \ --language en # Use a different model faster-whisper audio.wav \ --model /opt/faster-whisper/models/small.en \ --device cpu \ --language en # Compare speed between models using time time faster-whisper audio.wav \ --model /opt/faster-whisper/models/tiny.en time faster-whisper audio.wav \ --model /opt/faster-whisper/models/base.en time faster-whisper audio.wav \ --model /opt/faster-whisper/models/small.en
Commercial Cloud Alternatives
These companies are not sponsors or affiliated with this training or Call Center Village. They're listed for educational awareness only.
That being said, if any of you are reading this — Call Center Village is always looking for sponsors!
That being said, if any of you are reading this — Call Center Village is always looking for sponsors!
Commercial streaming STT. Purpose-built for real-time voice agents. Low latency with word-level timestamps.
Streaming STT with speaker diarization, sentiment analysis, and entity detection.
Known for TTS, also offers speech-to-text with low latency and multilingual support.
AI voice platform with speech-to-text capabilities alongside their voice generation tools.
Have a suggestion for a speech-to-text service to include here? Email us at [email protected]
Check Your Understanding
Question 1 / 3
By default, what does Faster Whisper do the first time you use a model?