tts.api

# From CRAN (once available)
install.packages("tts.api")

# Development version
remotes::install_github("cornball-ai/tts.api")

git clone https://github.com/cornball-ai/chatterbox-tts-api.git
cd chatterbox-tts-api

# For newer Nvidia GPUs (Blackwell/50xx series)
docker build -f docker/Dockerfile.blackwell -t chatterbox-tts:blackwell .

docker run -d \
  --name chatterbox-blackwell \
  --gpus all \
  -p 7810:4123 \
  -v $(pwd)/cache:/cache \
  -v $(pwd)/voices:/voices \
  chatterbox-tts:blackwell

docker run -d --gpus all --network=host --name qwen3-tts-api \
  -v ~/.cache/huggingface:/cache \
  -e PORT=7811 \
  -e USE_FLASH_ATTENTION=false \
  qwen3-tts-api:blackwell

library(tts.api)

# For local Chatterbox server (OpenAI-compatible)
set_tts_base("http://localhost:7810")

# For local Qwen3-TTS server
set_tts_base("http://localhost:7811")

# For OpenAI
set_tts_base("https://api.openai.com")
set_tts_key(Sys.getenv("OPENAI_API_KEY"))

# For ElevenLabs (separate API key)
set_elevenlabs_key(Sys.getenv("ELEVENLABS_API_KEY"))

tts_health()
#> $ok
#> [1] TRUE
#>
#> $status
#> [1] "OK (/health)"

voices()

# Basic usage (uses configured base URL)
tts(
  input = "Hello, world!",
  voice = "alloy",
  file = "hello.mp3"
)

# OpenAI with voice instructions
tts(
  input = "Today is a wonderful day to build something people love!",
  voice = "coral",
  file = "speech.mp3",
  backend = "openai",
  model = "gpt-4o-mini-tts",
  instructions = "Speak in a cheerful and positive tone."
)

# Chatterbox with custom parameters
tts(
  input = "Hello with my custom voice!",
  voice = "MyCustomVoice",
  file = "speech.wav",
  temperature = 0.9,
  exaggeration = 1.2,
  cfg_weight = 0.3
)

# ElevenLabs (different API, not OpenAI-compatible)
tts(
  input = "Hello from ElevenLabs!",
  voice = "21m00Tcm4TlvDq8ikWAM",  # Rachel voice ID
  file = "hello_eleven.mp3",
  backend = "elevenlabs",
  stability = 0.5,
  similarity_boost = 0.75
)

# Qwen3-TTS with built-in voice
tts(
  input = "Hello from Qwen3!",
  voice = "Vivian",
  file = "hello_qwen3.wav",
  backend = "qwen3"
)

# Return raw bytes (useful for Shiny)
audio_bytes <- tts(
  input = "Hello!",
  voice = "alloy"
)

# Fast mode (x-vector only, no transcript needed)
speech_clone(
  input = "Hello in my cloned voice!",
  voice_file = "reference.wav",
  x_vector_only = TRUE,
  file = "cloned.wav",
  backend = "qwen3"
)

# High quality mode (with transcript)
speech_clone(
  input = "Hello in my cloned voice!",
  voice_file = "reference.wav",
  ref_text = "This is what I said in the recording.",
  file = "cloned.wav",
  backend = "qwen3"
)

speech_design(
  input = "Hello, I am your AI assistant.",
  voice_description = "A warm, professional female voice with a slight British accent",
  file = "designed_voice.wav"
)

# Upload once
voice_upload(
  voice_file = "my_voice.wav",
  voice_name = "my-custom-voice"
)

# With language
voice_upload(
  voice_file = "french_voice.wav",
  voice_name = "french-speaker",
  language = "fr"
)

# Use the saved voice by name
tts(
  input = "Hello with my custom voice!",
  voice = "my-custom-voice",
  file = "output.wav"
)

# Or for one-off cloning (uploads and generates in one call)
speech_clone(
  input = "Hello with my custom voice!",
  voice_file = "my_voice.mp3",
  file = "output.wav",
  exaggeration = 0.8
)