llm-audio

Whisper transcription, translation, and TTS in a single C++ header.

Features

• Transcribe audio files (mp3, wav, m4a, ogg, webm) via OpenAI Whisper
• Translate audio to English
• Text-to-speech with 6 voices and 4 formats
• Single-header, C++17, namespace llm

Quick Start

#define LLM_AUDIO_IMPLEMENTATION
#include "llm_audio.hpp"

llm::TranscribeConfig cfg{ .api_key = "sk-..." };
auto result = llm::transcribe("audio.mp3", cfg);
std::cout << result.text;

API

TranscribeResult transcribe(const std::string& filepath, const TranscribeConfig&);
TranscribeResult transcribe_bytes(const std::vector<uint8_t>&, const std::string& filename, const TranscribeConfig&);
void             text_to_speech(const std::string& text, const std::string& output_path, const TTSConfig&);
std::vector<uint8_t> text_to_speech_bytes(const std::string& text, const TTSConfig&);

Build

cmake -B build && cmake --build build

Requires libcurl (vcpkg: vcpkg install curl).

License

MIT — Mattbusel, 2026

See Also

Repo	Purpose
llm-stream	SSE streaming
llm-cache	Response caching
llm-cost	Token cost estimation
llm-retry	Retry + circuit breaker
llm-format	Markdown/code formatting
llm-embed	Embeddings + cosine similarity
llm-pool	Connection pooling
llm-log	Structured logging
llm-template	Prompt templates
llm-agent	Tool-use agent loop
llm-rag	Retrieval-augmented generation
llm-eval	Output evaluation
llm-chat	Multi-turn chat
llm-vision	Vision/image inputs
llm-mock	Mock LLM for testing
llm-router	Model routing
llm-guard	Content moderation
llm-compress	Prompt compression
llm-batch	Batch processing
llm-audio	Audio transcription/TTS
llm-finetune	Fine-tuning jobs
llm-rank	Passage reranking
llm-parse	HTML/markdown parsing
llm-trace	Distributed tracing
llm-ab	A/B testing
llm-json	JSON parsing/building