Replicate voices

You can reproduce the voice, tone, cadence, and timbre of a speaker from a short audio sample using the voice replication feature. Both Gemini 3.8 Flash TTS (gemini-3.8-flash-tts) and Gemini 3.8 Flash-Lite TTS (gemini-3.8-flash-lite-tts) support voice replication.

To replicate a voice, you send a reference recording of the speaker (source_audio) and a recording of the same speaker reading a consent statement (consent_audio) to the Voices API. The service verifies consent and returns an identifier for the replicated voice. You then pass that identifier in speechConfig.voiceConfig.voice in generateContent or streamGenerateContent requests, without uploading the audio again.

Storage modes

When you create a replicated voice, the store field selects one of two storage modes:

  • Stored voice (store: true): Google stores the replicated voice in your project and returns a voice ID (voice_...) in the id field. You can list, get, and delete stored voices with the Voices API.
  • Voice replication key (store: false): Google doesn't store the replicated voice. The service returns an encrypted voice replication key (voicekey_...) in the key field, and your application stores the key. Use this mode if your workload requires that Google doesn't keep voice profiles.
Storage mode Identifier Management Expiration
Stored voice (store: true) voice_... Voices API list, get, and delete methods 1 year after last use
Voice replication key (store: false) voicekey_... Stored by your application. Google doesn't retain the key. 7 days after creation

Generating speech with a stored voice restarts its one-year retention period, so a voice that you use regularly doesn't expire. A request that uses an expired voice replication key fails with an INVALID_ARGUMENT error, and a request that uses a deleted or expired stored voice fails with a NOT_FOUND error. To keep using a replicated voice, create it again before it expires.

Before you begin

Complete the steps in Before you begin on the Gemini TTS page. The Voices API is available in the global location through the v1beta1 API version.

Audio and consent requirements

Each request to create a replicated voice needs two recordings of the same adult speaker:

  • Reference audio (source_audio): 10 to 30 seconds of clean, natural speech from the speaker whose voice you want to replicate. Longer samples generally produce better results.

  • Consent audio (consent_audio): A recording of the same speaker reading the consent statement for their language word for word. For the statements, see Supported languages and consent statements. In English, the statement is:

    "I am the owner of this voice and have consented to the creation of a synthetic model of my voice through the use of Google Cloud."

Both recordings must be WAV files with 16-bit little-endian linear PCM, mono, at 24 kHz.

If the consent recording doesn't match the consent statement, or the audio doesn't meet the usage guidelines, the create request fails with an INVALID_ARGUMENT error that describes the problem.

Create a stored replicated voice

Set the following fields in the request:

  • store: true.
  • voice.type: VOICE_TYPE_REPLICATED.
  • voice.display_name: Optional. A name for the voice.
  • voice.replicated.source_audio and voice.replicated.consent_audio: The base64-encoded reference and consent recordings, each with mime_type set to audio/wav.

Don't set voice.model. Requests that set a model together with store: true fail with an INVALID_ARGUMENT error.

The response returns the voice ID in the id field.

Python

import base64

from google import genai

client = genai.Client(enterprise=True, project="PROJECT_ID", location="global")

with open("SOURCE_AUDIO_PATH", "rb") as f:
    source_b64 = base64.b64encode(f.read()).decode("utf-8")

with open("CONSENT_AUDIO_PATH", "rb") as f:
    consent_b64 = base64.b64encode(f.read()).decode("utf-8")

replicated_voice = client.voices.create(
    store=True,
    voice={
        "type": "VOICE_TYPE_REPLICATED",
        "display_name": "Custom replicated speaker",
        "replicated": {
            "source_audio": {"mime_type": "audio/wav", "data": source_b64},
            "consent_audio": {"mime_type": "audio/wav", "data": consent_b64},
        },
    },
    timeout=60,
)

# The voice ID starts with "voice_".
print(f"Created voice ID: {replicated_voice.id}")

REST

SOURCE_B64=$(base64 -w 0 SOURCE_AUDIO_PATH)
CONSENT_B64=$(base64 -w 0 CONSENT_AUDIO_PATH)

cat > request.json <<EOF
{
  "store": true,
  "voice": {
    "type": "VOICE_TYPE_REPLICATED",
    "displayName": "Custom replicated speaker",
    "replicated": {
      "sourceAudio": {"mimeType": "audio/wav", "data": "${SOURCE_B64}"},
      "consentAudio": {"mimeType": "audio/wav", "data": "${CONSENT_B64}"}
    }
  }
}
EOF

curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  https://aiplatform.googleapis.com/v1beta1/projects/PROJECT_ID/locations/global/voices \
  -d @request.json | jq -r '.id'

Replace the following:

  • PROJECT_ID: Your Google Cloud project ID.
  • SOURCE_AUDIO_PATH: The path to the reference WAV file.
  • CONSENT_AUDIO_PATH: The path to the consent WAV file.

To list, get, or delete stored replicated voices, use the same methods as for designed voices. For details, see Manage stored voices.

Create a voice replication key

Set the following fields in the request:

  • store: false.
  • voice.type: VOICE_TYPE_REPLICATED.
  • voice.model: Required. The Gemini 3.8 TTS model to create the key with, for example gemini-3.8-flash-tts. The key works with both Gemini 3.8 TTS models.
  • voice.replicated.source_audio and voice.replicated.consent_audio: The base64-encoded reference and consent recordings, each with mime_type set to audio/wav.

The response returns the voice replication key in the key field.

Python

import base64

from google import genai

client = genai.Client(enterprise=True, project="PROJECT_ID", location="global")

with open("SOURCE_AUDIO_PATH", "rb") as f:
    source_b64 = base64.b64encode(f.read()).decode("utf-8")

with open("CONSENT_AUDIO_PATH", "rb") as f:
    consent_b64 = base64.b64encode(f.read()).decode("utf-8")

replicated_voice = client.voices.create(
    store=False,
    voice={
        "model": "gemini-3.8-flash-tts",
        "type": "VOICE_TYPE_REPLICATED",
        "replicated": {
            "source_audio": {"mime_type": "audio/wav", "data": source_b64},
            "consent_audio": {"mime_type": "audio/wav", "data": consent_b64},
        },
    },
    timeout=60,
)

# Store the key in your application. It starts with "voicekey_".
voice_key = replicated_voice.key

REST

SOURCE_B64=$(base64 -w 0 SOURCE_AUDIO_PATH)
CONSENT_B64=$(base64 -w 0 CONSENT_AUDIO_PATH)

cat > request.json <<EOF
{
  "store": false,
  "voice": {
    "model": "gemini-3.8-flash-tts",
    "type": "VOICE_TYPE_REPLICATED",
    "replicated": {
      "sourceAudio": {"mimeType": "audio/wav", "data": "${SOURCE_B64}"},
      "consentAudio": {"mimeType": "audio/wav", "data": "${CONSENT_B64}"}
    }
  }
}
EOF

curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  https://aiplatform.googleapis.com/v1beta1/projects/PROJECT_ID/locations/global/voices \
  -d @request.json | jq -r '.key'

Generate speech with a replicated voice

Pass the voice ID or the voice replication key in speechConfig.voiceConfig.voice:

Python

from google import genai

client = genai.Client(enterprise=True, project="PROJECT_ID", location="global")

response = client.models.generate_content(
    model="gemini-3.8-flash-tts",
    contents=[{
        "role": "user",
        "parts": [{
            "text": "Hello! This audio was generated with a replicated voice.",
            "speech_metadata": {"style": "warm and conversational"},
        }],
    }],
    config={
        "response_modalities": ["AUDIO"],
        "speech_config": {"voice_config": {"voice": "VOICE"}},
    },
)

# The SDK has already decoded the base64 audio, so inline_data.data is a
# complete WAV file by default.
with open("replicated_voice.wav", "wb") as f:
    f.write(response.candidates[0].content.parts[0].inline_data.data)

REST

curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  https://aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/global/publishers/google/models/gemini-3.8-flash-tts:generateContent \
  -d '{
    "contents": [{
      "role": "user",
      "parts": [{
        "text": "Hello! This audio was generated with a replicated voice.",
        "speechMetadata": {"style": "warm and conversational"}
      }]
    }],
    "generationConfig": {
      "responseModalities": ["AUDIO"],
      "speechConfig": {
        "voiceConfig": {"voice": "VOICE"}
      }
    }
  }' | jq -r '.candidates[0].content.parts[0].inlineData.data' | base64 --decode > replicated_voice.wav

Replace VOICE with the voice ID (voice_...) or the voice replication key (voicekey_...) that the create method returned.

Voice creation requests are subject to a per-project, per-minute quota. If you exceed it, the create method returns a RESOURCE_EXHAUSTED error. For more information, see Quotas and system limits.

The Voices API verifies consent by matching the transcript of consent_audio against the consent statement. The speaker must read the statement for their language word for word:

Language BCP-47 code Consent statement
Afrikaans af Ek is die eienaar van hierdie stem en het ingestem tot die skep van 'n sintetiese model van my stem deur die gebruik van Google Cloud.
Albanian sq Unë jam pronari i këtij zëri dhe kam dhënë pëlqimin për krijimin e një modeli sintetik të zërit tim përmes përdorimit të Google Cloud.
Amharic am የዚህ ድምጽ ባለቤት እኔ ነኝ እና የጉግል ክላውድን በመጠቀም የድምፄ ሰው ሰራሽ ሞዴል እንዲፈጠር ተስማምቻለሁ።
Arabic ar أنا صاحب هذا الصوت وقد وافقت على إنشاء نموذج اصطناعي لصوتي باستخدام خدمة جوجل كلاود.
Armenian hy Ես այս ձայնի սեփականատերն եմ և համաձայնություն եմ տվել Google Cloud-ի միջոցով իմ ձայնի սինթետիկ մոդելի ստեղծմանը։
Azerbaijani az Mən bu səsin sahibiyəm və Google Cloud vasitəsilə səsimin sintetik modelinin yaradılmasına razılıq vermişəm.
Bangla bn আমি এই কণ্ঠস্বরের মালিক এবং গুগল ক্লাউড ব্যবহারের মাধ্যমে আমার কণ্ঠস্বরের একটি কৃত্রিম মডেল তৈরিতে সম্মতি দিয়েছি।
Basque eu Ahots honen jabea naiz eta nire ahotsaren eredu sintetiko bat sortzeko baimena eman dut Google Cloud erabiliz.
Belarusian be Я з'яўляюся ўладальнікам гэтага голасу і даў згоду на стварэнне сінтэтычнай мадэлі майго голасу з дапамогай Google Cloud
Bulgarian bg Аз съм собственикът на този глас и съм дал съгласието си за създаването на синтетичен модел на моя глас чрез използването на Google Cloud.
Burmese my ကျွန်ုပ်သည် ဤအသံ၏ပိုင်ရှင်ဖြစ်ပြီး Google Cloud ကိုအသုံးပြုခြင်းဖြင့် ကျွန်ုပ်၏အသံ၏ ပေါင်းစပ်ပုံစံတစ်ခု ဖန်တီးရန် သဘောတူပါသည်။
Catalan ca Sóc el propietari d'aquesta veu i he donat el meu consentiment per a la creació d'un model sintètic de la meva veu mitjançant l'ús de Google Cloud.
Cebuano ceb Ako ang tag-iya niining tingog ug miuyon sa paghimo og sintetikong modelo sa akong tingog pinaagi sa paggamit sa Google Cloud.
Chinese, Mandarin cmn 我是这段声音的所有者，并已同意通过 Google Cloud 创建我的声音合成模型。
Croatian hr Vlasnik sam ovog glasa i pristao sam na izradu sintetičkog modela mog glasa putem Google Clouda.
Czech cs Jsem vlastníkem tohoto hlasu a souhlasil(a) jsem s vytvořením syntetického modelu mého hlasu pomocí služby Google Cloud.
Danish da Jeg ejer denne stemme og har givet samtykke til oprettelsen af en syntetisk model af min stemme ved hjælp af Google Cloud.
Dutch nl Ik ben de eigenaar van deze stem en heb toestemming gegeven voor het creëren van een synthetisch model van mijn stem via Google Cloud.
English en I am the owner of this voice and have consented to the creation of a synthetic model of my voice through the use of Google Cloud
Estonian et Olen selle hääle omanik ja olen andnud nõusoleku oma hääle sünteetilise mudeli loomiseks Google Cloudi abil.
Filipino fil Ako ang may-ari ng boses na ito at pumayag sa paglikha ng isang sintetikong modelo ng aking boses sa pamamagitan ng paggamit ng Google Cloud.
Finnish fi Olen tämän äänen omistaja ja olen antanut suostumukseni synteettisen ääneni mallintamiseen Google Cloudin avulla.
French fr Je suis propriétaire de cette voix et j'ai consenti à la création d'un modèle synthétique de ma voix grâce à l'utilisation de Google Cloud.
Galician gl Son o propietario desta voz e dei o meu consentimento para a creación dun modelo sintético da miña voz mediante o uso de Google Cloud.
Georgian ka მე ვარ ამ ხმის მფლობელი და დავთანხმდი ჩემი ხმის სინთეზური მოდელის შექმნას Google Cloud-ის გამოყენებით.
German de Ich bin der Inhaber dieser Stimme und habe der Erstellung eines synthetischen Modells meiner Stimme mithilfe von Google Cloud zugestimmt.
Greek el Είμαι ο κάτοχος αυτής της φωνής και έχω συναινέσει στη δημιουργία ενός συνθετικού μοντέλου της φωνής μου μέσω της χρήσης του Google Cloud.
Gujarati gu હું આ અવાજનો માલિક છું અને ગૂગલ ક્લાઉડના ઉપયોગ દ્વારા મારા અવાજનું સિન્થેટિક મોડેલ બનાવવા માટે સંમતિ આપી છે.
Haitian Creole ht Mwen se pwopriyetè vwa sa a epi mwen bay konsantman pou kreyasyon yon modèl sentetik vwa mwen atravè itilizasyon Google Cloud.
Hebrew he אני הבעלים של קול זה והסכמתי ליצירת מודל סינתטי של קולי באמצעות Google Cloud.
Hindi hi मैं इस आवाज का स्वामी हूं और मैंने Google क्लाउड के उपयोग के माध्यम से अपनी आवाज का कृत्रिम मॉडल बनाने की सहमति दी है।
Hungarian hu Én vagyok a hang tulajdonosa, és hozzájárultam ahhoz, hogy a Google Cloud segítségével szintetikus modellt készítsenek a hangomról.
Icelandic is Ég er eigandi þessarar raddar og hef samþykkt að búið sé til tilbúið líkan af rödd minni með því að nota Google Cloud.
Indonesian id Saya adalah pemilik suara ini dan telah menyetujui pembuatan model sintetis suara saya melalui penggunaan Google Cloud.
Italian it Sono il proprietario di questa voce e ho acconsentito alla creazione di un modello sintetico della mia voce tramite l'utilizzo di Google Cloud.
Japanese ja 私はこの音声の所有者であり、Google Cloud を利用して私の音声の合成モデルを作成することに同意しました。
Javanese jv Aku sing nduwèni swara iki lan wis sarujuk kanggo nggawé model sintetis swaraku liwat panggunaan Google Cloud.
Kannada kn ನಾನು ಈ ಧ್ವನಿಯ ಮಾಲೀಕ ಮತ್ತು Google Cloud ಬಳಕೆಯ ಮೂಲಕ ನನ್ನ ಧ್ವನಿಯ ಸಂಶ್ಲೇಷಿತ ಮಾದರಿಯನ್ನು ರಚಿಸಲು ಸಮ್ಮತಿಸಿದ್ದೇನೆ.
Korean ko 저는 이 음성의 소유자이며, Google Cloud를 사용하여 제 음성의 합성 모델을 생성하는 데 동의했습니다.
Lao lo ຂ້ອຍເປັນເຈົ້າຂອງສຽງນີ້ ແລະ ໄດ້ຍິນຍອມໃຫ້ສ້າງຮູບແບບສັງເຄາະສຽງຂອງຂ້ອຍຜ່ານການນຳໃຊ້ Google Cloud
Latin la Huius vocis dominus sum et consensi sum ad creandum exemplar syntheticum vocis meae per usum Google Cloud.
Latvian lv Esmu šīs balss īpašnieks un esmu piekritis sintētiska manas balss modeļa izveidei, izmantojot Google Cloud.
Lithuanian lt Esu šio balso savininkas ir sutikau, kad naudojant „Google Cloud“ būtų sukurtas sintetinis mano balso modelis.
Luxembourgish lb Ech sinn de Besëtzer vun dëser Stëmm an hunn der Erstellung vun engem synthetesche Modell vu menger Stëmm mat Hëllef vu Google Cloud zougestëmmt.
Macedonian mk Јас сум сопственик на овој глас и се согласив за создавање синтетички модел на мојот глас преку користење на Google Cloud.
Malagasy mg Izaho no tompon'ity feo ity ary nanaiky ny hamorona modely sentetika amin'ny feoko amin'ny alàlan'ny fampiasana ny Google Cloud.
Malay ms Saya pemilik suara ini dan telah bersetuju untuk penciptaan model sintetik suara saya melalui penggunaan Google Cloud.
Malayalam ml ഈ ശബ്ദത്തിന്റെ ഉടമ ഞാനാണ്, Google ക്ലൗഡ് ഉപയോഗിച്ച് എന്റെ ശബ്ദത്തിന്റെ ഒരു സിന്തറ്റിക് മോഡൽ സൃഷ്ടിക്കാൻ ഞാൻ സമ്മതം നൽകിയിട്ടുണ്ട്.
Marathi mr मी या आवाजाचा मालक आहे आणि गूगल क्लाउडचा वापर करून माझ्या आवाजाचे कृत्रिम मॉडेल तयार करण्यास मी संमती दिली आहे.
Mongolian mn Би энэ дуу хоолойны эзэмшигч бөгөөд Google Cloud ашиглан өөрийн дуу хоолойны синтетик загварыг бүтээхийг зөвшөөрсөн.
Nepali ne म यो आवाजको मालिक हुँ र गुगल क्लाउडको प्रयोग मार्फत मेरो आवाजको सिंथेटिक मोडेल सिर्जना गर्न सहमति दिएको छु।
Norwegian, Bokmål nb Jeg eier denne stemmen og har samtykket til at det opprettes en syntetisk modell av stemmen min ved bruk av Google Cloud.
Odia or ମୁଁ ଏହି ଭଏସ୍‌ର ମାଲିକ ଏବଂ Google Cloud ବ୍ୟବହାର ମାଧ୍ୟମରେ ମୋ ଭଏସ୍‌ର ଏକ ସିନ୍ଥେଟିକ୍ ମଡେଲ୍ ତିଆରି କରିବାକୁ ସମ୍ମତି ଦେଇଛି।
Pashto ps زه د دې غږ مالک یم او د ګوګل کلاوډ په کارولو سره مې د خپل غږ د مصنوعي ماډل جوړولو ته رضایت ورکړی دی.
Persian fa من مالک این صدا هستم و با ایجاد یک مدل مصنوعی از صدایم از طریق استفاده از Google Cloud موافقت کرده‌ام.
Polish pl Jestem właścicielem tego głosu i wyraziłem zgodę na utworzenie syntetycznego modelu mojego głosu za pomocą Google Cloud
Portuguese pt Eu sou o proprietário desta voz e consinto com a criação de um modelo sintético da minha voz através do uso do Google Cloud.
Punjabi pa ਮੈਂ ਇਸ ਆਵਾਜ਼ ਦਾ ਮਾਲਕ ਹਾਂ ਅਤੇ ਗੂਗਲ ਕਲਾਉਡ ਦੀ ਵਰਤੋਂ ਰਾਹੀਂ ਆਪਣੀ ਆਵਾਜ਼ ਦੇ ਇੱਕ ਸਿੰਥੈਟਿਕ ਮਾਡਲ ਨੂੰ ਬਣਾਉਣ ਲਈ ਸਹਿਮਤੀ ਦਿੱਤੀ ਹੈ।
Romanian ro Sunt proprietarul acestei voci și am consimțit la crearea unui model sintetic al vocii mele prin utilizarea Google Cloud.
Russian ru Я являюсь владельцем этого голоса и дал согласие на создание синтетической модели моего голоса с использованием Google Cloud.
Serbian sr Ја сам власник овог гласа и пристао/ла сам на креирање синтетичког модела мог гласа коришћењем Google Cloud-а
Sindhi sd مان هن آواز جو مالڪ آهيان ۽ گوگل ڪلائوڊ جي استعمال ذريعي پنهنجي آواز جي هڪ مصنوعي ماڊل جي تخليق تي رضامندي ڏني آهي.
Sinhala si මම මෙම හඬෙහි හිමිකරු වන අතර Google Cloud භාවිතය හරහා මගේ හඬෙහි කෘතිම ආකෘතියක් නිර්මාණය කිරීමට කැමැත්ත පළ කර ඇත්තෙමි.
Slovak sk Som vlastníkom tohto hlasu a súhlasil/a som s vytvorením syntetického modelu môjho hlasu pomocou služby Google Cloud.
Slovenian sl Sem lastnik tega glasu in sem privolil v ustvarjanje sintetičnega modela mojega glasu z uporabo storitve Google Cloud.
Spanish es Soy el propietario de esta voz y he consentido la creación de un modelo sintético de mi voz mediante el uso de Google Cloud.
Swahili sw Mimi ndiye mmiliki wa sauti hii na nimekubali kuundwa kwa modeli ya sauti yangu iliyotengenezwa kwa njia ya Google Cloud.
Swedish sv Jag äger den här rösten och har samtyckt till att en syntetisk modell av min röst skapas med hjälp av Google Cloud.
Tamil ta இந்தக் குரலின் உரிமையாளர் நானே, மேலும் கூகிள் கிளவுடைப் பயன்படுத்தி எனது குரலின் செயற்கை மாதிரியை உருவாக்குவதற்கும் நான் ஒப்புதல் அளித்துள்ளேன்.
Telugu te ఈ స్వరానికి నేను యజమానిని మరియు గూగుల్ క్లౌడ్‌ను ఉపయోగించి నా స్వరం యొక్క కృత్రిమ నమూనాను రూపొందించడానికి నేను సమ్మతించాను.
Thai th ฉันเป็นเจ้าของเสียงนี้และได้ยินยอมให้สร้างแบบจำลองเสียงสังเคราะห์ของเสียงฉันโดยใช้ Google Cloud
Turkish tr Bu sesin sahibi benim ve Google Cloud kullanılarak sesimin sentetik bir modelinin oluşturulmasına izin verdim.
Ukrainian uk Я є власником цього голосу та дав згоду на створення синтетичної моделі мого голосу за допомогою Google Cloud
Urdu ur میں اس آواز کا مالک ہوں اور میں نے گوگل کلاؤڈ کے استعمال کے ذریعے اپنی آواز کا مصنوعی ماڈل بنانے پر رضامندی دی ہے۔
Vietnamese vi Tôi là chủ sở hữu giọng nói này và đã đồng ý cho phép tạo ra một mô hình tổng hợp giọng nói của mình thông qua việc sử dụng Google Cloud.

Best practices

  • Record in a quiet environment: Minimize echo, background noise, music, and overlapping voices. Output quality depends on the quality of source_audio.
  • Match recording conditions: Record source_audio and consent_audio with the same microphone in the same room so that speaker verification succeeds.
  • Convert to the required format: Resample recordings to 24 kHz, mono, 16-bit PCM WAV before you encode them.
  • Plan for renewal: Voice replication keys are valid for seven days, and stored voices expire one year after they were last used. Keep the source and consent recordings, or a process to collect them, so that you can create the voice again before it expires.

What's next