feat(clients): google cloud voice: generic client, TTS, STT, Gemini audio, live voice page

src/google: GoogleClient over molo::clients (key in x-goog-api-key, never in
URLs), Service base, TextToSpeech, SpeechToText, Generative (replies, audio
in, audio out), Recognizer/Synthesizer seams with cloud, gemini and vertex
backends, VoicePipeline with bounded history. Example: /voice live
conversation page, /voice/{say,hear,turn,reset}, /google/status. Keys load
from env, .env.json.local, .env.json. Paid routes are excluded from locust.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-10-24 04:05:41 +02:00
co-authored by Claude Sonnet 5.5
parent dd773d3015
commit 837de58136
54 changed files with 3086 additions and 21 deletions
+65
View File
@@ -0,0 +1,65 @@
#include "google/text_to_speech.h"
#include "google/caps.h"
#include "google/endpoints.h"
#include "security/crypto/Base64.h"
namespace molo::google {
using clients::body::Json;
TextToSpeech::TextToSpeech(std::shared_ptr<GoogleClient> client)
: Service(std::move(client), endpoints::text_to_speech()) {}
Audio TextToSpeech::synthesize(const SynthesisRequest& request) const {
if (request.text.empty()) throw GoogleError::invalid("text is empty");
if (request.text.size() > Caps::kMaxTtsChars)
throw GoogleError::invalid("text over " +
std::to_string(Caps::kMaxTtsChars) + " characters");
if (request.speaking_rate < 0.25 || request.speaking_rate > 4.0)
throw GoogleError::invalid("speaking_rate outside 0.25..4.0");
Json input = Json::object();
input.set(request.ssml ? "ssml" : "text", Json(request.text));
Json voice = Json::object();
voice.set("languageCode", Json(request.language));
if (!request.voice.empty()) voice.set("name", Json(request.voice));
Json audio_config = Json::object();
audio_config.set("audioEncoding", Json(audio_encoding_name(request.encoding)));
audio_config.set("speakingRate", Json(request.speaking_rate));
Json body = Json::object();
body.set("input", input);
body.set("voice", voice);
body.set("audioConfig", audio_config);
Json reply = call("/text:synthesize", body);
Audio out;
out.mime = audio_encoding_mime(request.encoding);
if (!security::crypto::Base64::decode(string_at(reply, "audioContent"),
out.bytes) ||
out.bytes.empty())
throw GoogleError(502, "BAD_RESPONSE", "missing or invalid audioContent",
endpoint_.name);
return out;
}
std::vector<Voice> TextToSpeech::voices(const std::string& language) const {
Json reply = fetch(language.empty() ? "/voices"
: "/voices?languageCode=" + language);
std::vector<Voice> out;
const Json* list = reply.find("voices");
if (!list || !list->is_array()) return out;
for (const Json& v : list->as_array()) {
Voice voice;
voice.name = string_at(v, "name");
voice.gender = string_at(v, "ssmlGender");
if (const Json* codes = v.find("languageCodes");
codes && codes->is_array())
for (const Json& c : codes->as_array())
if (c.is_string()) voice.languages.push_back(c.as_string());
out.push_back(std::move(voice));
}
return out;
}
} // namespace molo::google