OmniVoice HTTP TTS Server

CMakeLists.txt

@@ -4,6 +4,8 @@ project(omnivoice-ggml LANGUAGES C CXX)
 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 
+option(OV_WEBSERVER "build webserver" ON)
+
 # version.h: embed git commit hash into all binaries.
 # runs on every build, only rewrites if the hash changed.
 set(VERSION_OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/version.h")
@@ -139,6 +141,16 @@ add_executable(omnivoice-tts tools/omnivoice-tts.cpp)
 target_link_libraries(omnivoice-tts PRIVATE omnivoice-core)
 link_ggml_backends(omnivoice-tts)
 
+# omnivoice-tts-server : an HTTP webserver backed by the TTS pipeline
+if(OV_WEBSERVER)
+    find_package(httplib REQUIRED)
+    find_package(nlohmann_json REQUIRED)
+
+    add_executable(omnivoice-tts-server tools/omnivoice-tts-server.cpp)
+    target_link_libraries(omnivoice-tts-server PRIVATE omnivoice-core httplib::httplib nlohmann_json::nlohmann_json)
+    link_ggml_backends(omnivoice-tts-server)
+endif()
+
 # test-abi-c : pure C99 smoke test that locks in the public ABI contract.
 # Compiles omnivoice.h with a C compiler under -Wall -Werror -pedantic and
 # links against the static lib. The test never loads a model ; failure

examples/client.sh

@@ -0,0 +1,32 @@
+#!/bin/bash
+
+set -eu
+
+if ! command -v jq; then
+  echo "This script requires 'jq', please install it before you proceed"
+  exit 1
+fi
+
+if [ $# -lt 1 ]; then
+  echo "Usage $0 input [lang format instructions]"
+  echo "Environment variables:"
+  echo
+  echo -e "OMNIVOICE_SERVER\tHTTP server URL (default: http://127.0.0.1:1234)"
+  echo -e "OMNIVOICE_OUTPUT\tOutput path (default: output.wav)"
+  echo -e "OV_TIMEOUT\tClient request timeout in seconds (default: 300)"
+  exit 255
+fi
+
+: "${OV_SERVER:=http://127.0.0.1:1234}"
+: "${OV_OUTPUT:=output.wav}"
+: "${OV_TIMEOUT:=300}"
+PAYLOAD=$(jq -cn --arg input "$1" --arg lang "${2:-}" --arg format "${3:-}" --arg instructions "${4:-}" \
+  'if $input != "" then {input: $input} else . end +
+   if $lang != "" then {lang: $lang} else . end +
+   if $format != "" then {format: $format} else . end +
+   if $instructions != "" then {instructions: $instructions} else . end')
+
+exec curl -v --max-time $OV_TIMEOUT --request POST \
+  --header "Content-Type: application/json" \
+  --data "$PAYLOAD" "$OV_SERVER"/v1/audio/speech \
+  --output "$OV_OUTPUT"

examples/server.sh

@@ -0,0 +1,10 @@
+#!/bin/bash
+
+set -eu
+
+exec ../build/omnivoice-tts-server \
+    --listen-ip 127.0.0.1 \
+    --model ../models/omnivoice-base-Q8_0.gguf \
+    --codec ../models/omnivoice-tokenizer-Q8_0.gguf \
+    --instruct "male, young adult, moderate pitch" \
+    --lang English "$@"