Add voice library for ~/.cornball/voices/

NAMESPACE

@@ -17,5 +17,8 @@ export(tts)
 export(tts_health)
 export(tts_providers)
 export(tts_voices)
+export(voice_ensure)
+export(voice_file)
+export(voice_library)
 export(voice_upload)
 export(voices)

R/gpuctl_integration.R

@@ -0,0 +1,92 @@
+# gpu.ctl integration for tts.api
+#
+# Optionally acquires GPU resources before API calls when gpu.ctl is available.
+# Enable with: options(tts.gpuctl = TRUE)
+
+# Service registry for TTS backends
+.tts_gpu_services <- list(
+    chatterbox = list(
+        name = "chatterbox",
+        port = 7810,
+        vram = 6,
+        container = "chatterbox",
+        health = "/health"
+    ),
+    qwen3 = list(
+        name = "qwen3-tts",
+        port = 7811,
+        vram = 8,
+        container = "qwen3-tts-api",
+        health = "/health"
+    )
+)
+
+#' Check if gpu.ctl integration is enabled
+#' @noRd
+.gpuctl_enabled <- function () {
+    isTRUE(getOption("tts.gpuctl", FALSE)) &&
+    requireNamespace("gpu.ctl", quietly = TRUE)
+}
+
+#' Register tts.api service with gpu.ctl
+#' @param backend Character. Backend name ("chatterbox" or "qwen3")
+#' @noRd
+.gpuctl_register_service <- function (backend = "chatterbox") {
+    if (!.gpuctl_enabled()) return(invisible(FALSE))
+
+    service <- .tts_gpu_services[[backend]]
+    if (is.null(service)) {
+        warning("Unknown backend for gpu.ctl: ", backend, call. = FALSE)
+        return(invisible(FALSE))
+    }
+
+    tryCatch({
+            # Only register if not already registered
+            existing <- gpu.ctl::gpu_services()
+            if (!service$name %in% existing$name) {
+                gpu.ctl::gpu_register(
+                    name = service$name,
+                    port = service$port,
+                    vram = service$vram,
+                    container = service$container,
+                    health_endpoint = service$health
+                )
+            }
+        }, error = function (e) {
+            # Silently ignore registration errors
+        })
+    invisible(TRUE)
+}
+
+#' Acquire GPU for TTS backend if gpu.ctl is enabled
+#'
+#' @param backend Character. Backend name ("chatterbox" or "qwen3")
+#' @return Invisible TRUE if acquired, FALSE if not using gpu.ctl
+#' @noRd
+.gpuctl_acquire <- function (backend = "chatterbox") {
+    if (!.gpuctl_enabled()) return(invisible(FALSE))
+
+    service <- .tts_gpu_services[[backend]]
+    if (is.null(service)) {
+        warning("Unknown backend for gpu.ctl: ", backend, call. = FALSE)
+        return(invisible(FALSE))
+    }
+
+    .gpuctl_register_service(backend)
+
+    tryCatch({
+            gpu.ctl::gpu_acquire(service$name)
+
+            # Auto-set tts.api_base from gpu.ctl's service URL
+            url <- gpu.ctl::gpu_service_url(service$name)
+            if (!is.null(url)) {
+                options(tts.api_base = url)
+            }
+
+            invisible(TRUE)
+        }, error = function (e) {
+            warning("gpu.ctl: ", e$message, call. = FALSE)
+            invisible(FALSE)
+        })
+}
+

R/internal_chatterbox.R

@@ -93,7 +93,7 @@ clear_native_chatterbox_cache <- function () {
 
     # Generate speech (traced = TRUE for ~5x faster inference)
     result <- tryCatch(
-        chatterbox::tts(
+        chatterbox::generate(
             model = model,
             text = input,
             voice = voice,

R/tts.R

@@ -90,6 +90,11 @@ tts <- function (input, voice, file = NULL,
         }
     }
 
+    # Auto-acquire GPU for container backends via gpu.ctl
+    if (backend %in% c("chatterbox", "qwen3")) {
+        .gpuctl_acquire(backend)
+    }
+
     # Dispatch to native chatterbox package
     if (backend == "native") {
         return(.via_chatterbox(

R/tts_providers.R

@@ -106,15 +106,11 @@ tts_voices <- function (provider, base_url = NULL, timeout = 2) {
         }
     }
 
-    # For Chatterbox, try reading from filesystem as fallback
+    # For Chatterbox, try reading from voice library as fallback
     if (provider == "Chatterbox (Local)") {
-        voices_dir <- Sys.getenv("CHATTERBOX_VOICES_DIR", "~/chatterbox-tts-api/voices")
-        voices_dir <- path.expand(voices_dir)
-        if (dir.exists(voices_dir)) {
-            files <- list.files(voices_dir, pattern = "\\.(mp3|wav)$", ignore.case = TRUE)
-            if (length(files) > 0) {
-                return(tools::file_path_sans_ext(files))
-            }
+        lib <- voice_library()
+        if (length(lib) > 0) {
+            return(names(lib))
         }
     }
 

R/voice_library.R

@@ -0,0 +1,115 @@
+#' List Voice Files in Library
+#'
+#' Scans a directory for voice sample files and returns their paths.
+#'
+#' @param voices_dir Path to voice library directory.
+#'   Defaults to \code{TTS_VOICES_DIR} env var, then \code{~/.cornball/voices}.
+#' @return Named character vector: names are voice names (filename without
+#'   extension), values are full file paths. Returns empty named character
+#'   vector if directory doesn't exist or contains no voice files.
+#'
+#' @export
+#' @examples
+#' \dontrun{
+#' voice_library()
+#' voice_library("~/my-voices")
+#' }
+voice_library <- function(voices_dir = NULL) {
+    if (is.null(voices_dir)) {
+        voices_dir <- Sys.getenv("TTS_VOICES_DIR", "~/.cornball/voices")
+    }
+    voices_dir <- path.expand(voices_dir)
+
+    if (!dir.exists(voices_dir)) {
+        return(structure(character(0), names = character(0)))
+    }
+
+    files <- list.files(voices_dir, pattern = "\\.(wav|mp3|m4a|flac)$",
+                        full.names = TRUE, ignore.case = TRUE)
+    if (length(files) == 0) {
+        return(structure(character(0), names = character(0)))
+    }
+
+    names(files) <- tools::file_path_sans_ext(basename(files))
+    files
+}
+
+#' Resolve Voice Name to File Path
+#'
+#' Looks up a voice by name in the voice library directory.
+#' Case-insensitive matching is attempted if exact match fails.
+#'
+#' @param voice_name Character. Name of the voice to find.
+#' @param voices_dir Path to voice library directory (see \code{\link{voice_library}}).
+#' @return Character string: full path to the voice file.
+#'   Stops with an error if the voice is not found.
+#'
+#' @export
+#' @examples
+#' \dontrun{
+#' voice_file("BigCasey")
+#' voice_file("delicatecasey")
+#' }
+voice_file <- function(voice_name, voices_dir = NULL) {
+    lib <- voice_library(voices_dir)
+
+    # Exact match
+    if (voice_name %in% names(lib)) {
+        return(unname(lib[voice_name]))
+    }
+
+    # Case-insensitive match
+    idx <- match(tolower(voice_name), tolower(names(lib)))
+    if (!is.na(idx)) {
+        return(unname(lib[idx]))
+    }
+
+    stop("Voice '", voice_name, "' not found in library. Available: ",
+         paste(names(lib), collapse = ", "), call. = FALSE)
+}
+
+#' Ensure Voice is Available on TTS Server
+#'
+#' Checks if a voice exists on the current TTS server. If not, finds the
+#' voice file in the library and uploads it.
+#'
+#' @param voice_name Character. Name of the voice.
+#' @param voices_dir Path to voice library directory (see \code{\link{voice_library}}).
+#' @return Invisible \code{TRUE} if the voice is available (already present or
+#'   successfully uploaded).
+#'
+#' @export
+#' @examples
+#' \dontrun{
+#' set_tts_base("http://localhost:4123")
+#' voice_ensure("BigCasey")
+#' tts("Hello!", voice = "BigCasey", file = "out.mp3")
+#' }
+voice_ensure <- function(voice_name, voices_dir = NULL) {
+    # Check if voice already on server
+    server_voices <- tryCatch(voices(), error = function(e) NULL)
+
+    if (!is.null(server_voices)) {
+        # Normalize to character vector of names
+        v_names <- if (is.data.frame(server_voices) && "name" %in% names(server_voices)) {
+            server_voices$name
+        } else if (is.data.frame(server_voices) && "id" %in% names(server_voices)) {
+            server_voices$id
+        } else if (is.character(server_voices)) {
+            server_voices
+        } else {
+            character(0)
+        }
+
+        if (tolower(voice_name) %in% tolower(v_names)) {
+            message("Voice '", voice_name, "' already on server")
+            return(invisible(TRUE))
+        }
+    }
+
+    # Find and upload
+    vf <- voice_file(voice_name, voices_dir)
+    message("Uploading voice '", voice_name, "' from ", vf)
+    voice_upload(voice_file = vf, voice_name = voice_name, language = "en")
+    invisible(TRUE)
+}

man/dot-tts_elevenlabs.Rd

@@ -3,14 +3,8 @@
 \alias{.tts_elevenlabs}
 \title{ElevenLabs TTS backend}
 \usage{
-.tts_elevenlabs(
-  input,
-  voice_id,
-  file = NULL,
-  model = NULL,
-  stability = NULL,
-  similarity_boost = NULL
-)
+.tts_elevenlabs(input, voice_id, file = NULL, model = NULL, stability = NULL,
+                similarity_boost = NULL)
 }
 \description{
 ElevenLabs TTS backend

man/dot-via_chatterbox.Rd

@@ -3,15 +3,8 @@
 \alias{.via_chatterbox}
 \title{Internal: Generate speech via native chatterbox package}
 \usage{
-.via_chatterbox(
-  input,
-  voice,
-  file = NULL,
-  exaggeration = NULL,
-  cfg_weight = NULL,
-  temperature = NULL,
-  device = "cuda"
-)
+.via_chatterbox(input, voice, file = NULL, exaggeration = NULL,
+                cfg_weight = NULL, temperature = NULL, device = "cuda")
 }
 \arguments{
 \item{input}{Character. Text to convert to speech.}

man/elevenlabs_voice_upload.Rd

@@ -3,13 +3,8 @@
 \alias{elevenlabs_voice_upload}
 \title{Upload Voice to ElevenLabs (Instant Voice Clone)}
 \usage{
-elevenlabs_voice_upload(
-  files,
-  name,
-  description = NULL,
-  remove_background_noise = FALSE,
-  labels = NULL
-)
+elevenlabs_voice_upload(files, name, description = NULL,
+                        remove_background_noise = FALSE, labels = NULL)
 }
 \arguments{
 \item{files}{Character vector. Paths to audio files (1-25 files, each up to 10MB).}

man/speech_clone.Rd

@@ -3,20 +3,10 @@
 \alias{speech_clone}
 \title{Generate Speech with Voice Cloning}
 \usage{
-speech_clone(
-  input,
-  voice_file,
-  file = NULL,
-  backend = c("auto", "chatterbox", "qwen3"),
-  ref_text = NULL,
-  x_vector_only = FALSE,
-  language = NULL,
-  exaggeration = NULL,
-  temperature = NULL,
-  cfg_weight = NULL,
-  speed = NULL,
-  seed = NULL
-)
+speech_clone(input, voice_file, file = NULL,
+             backend = c("auto", "chatterbox", "qwen3"), ref_text = NULL,
+             x_vector_only = FALSE, language = NULL, exaggeration = NULL,
+             temperature = NULL, cfg_weight = NULL, speed = NULL, seed = NULL)
 }
 \arguments{
 \item{input}{Character. The text to convert to speech.}

man/tts.Rd

@@ -3,24 +3,12 @@
 \alias{tts}
 \title{Generate Speech from Text}
 \usage{
-tts(
-  input,
-  voice,
-  file = NULL,
-  backend = c("auto", "native", "chatterbox", "qwen3", "openai", "elevenlabs"),
-  model = NULL,
-  temperature = NULL,
-  speed = NULL,
-  exaggeration = NULL,
-  cfg_weight = NULL,
-  stability = NULL,
-  similarity_boost = NULL,
-  seed = NULL,
-  response_format = NULL,
-  instructions = NULL,
-  language = NULL,
-  device = "cuda"
-)
+tts(input, voice, file = NULL,
+    backend = c("auto", "native", "chatterbox", "qwen3", "openai", "elevenlabs"),
+    model = NULL, temperature = NULL, speed = NULL, exaggeration = NULL,
+    cfg_weight = NULL, stability = NULL, similarity_boost = NULL, seed = NULL,
+    response_format = NULL, instructions = NULL, language = NULL,
+    device = "cuda")
 }
 \arguments{
 \item{input}{Character. The text to convert to speech.}

man/voice_ensure.Rd

@@ -0,0 +1,27 @@
+% tinyrox says don't edit this manually, but it can't stop you!
+\name{voice_ensure}
+\alias{voice_ensure}
+\title{Ensure Voice is Available on TTS Server}
+\usage{
+voice_ensure(voice_name, voices_dir = NULL)
+}
+\arguments{
+\item{voice_name}{Character. Name of the voice.}
+
+\item{voices_dir}{Path to voice library directory (see \code{\link{voice_library}}).}
+}
+\value{
+Invisible \code{TRUE} if the voice is available (already present or
+  successfully uploaded).
+}
+\description{
+Checks if a voice exists on the current TTS server. If not, finds the
+voice file in the library and uploads it.
+}
+\examples{
+\dontrun{
+set_tts_base("http://localhost:4123")
+voice_ensure("BigCasey")
+tts("Hello!", voice = "BigCasey", file = "out.mp3")
+}
+}