{"id":2431,"date":"2026-06-09T09:00:00","date_gmt":"2026-06-09T09:00:00","guid":{"rendered":"https:\/\/wiro.ai\/blog\/?p=2431"},"modified":"2026-09-27T20:52:07","modified_gmt":"2026-09-27T20:52:07","slug":"chatterbox-multilingual-voice-cloning-in-23-languages","status":"publish","type":"post","link":"https:\/\/wiro.ai\/blog\/chatterbox-multilingual-voice-cloning-in-23-languages\/","title":{"rendered":"Chatterbox Multilingual: Voice Cloning in 23 Languages"},"content":{"rendered":"<p><strong>Chatterbox Multilingual voice cloning<\/strong> is useful only if a reference voice still feels usable when the script, pacing, and language change. This six-output test keeps one reference clip fixed, then changes language, delivery, and guidance settings. The point was not to crown a winner from a few clips. It was to make the controls and trade-offs concrete.<\/p>\n<figure><img loading=\"lazy\" decoding=\"async\" width=\"2528\" height=\"1696\" class=\"wp-image-2418\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-cover.jpg\" alt=\"Cover image for the Chatterbox Multilingual post\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-cover.jpg 2528w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-cover-510x342.jpg 510w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-cover-900x604.jpg 900w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-cover-768x515.jpg 768w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-cover-1536x1030.jpg 1536w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-cover-2048x1374.jpg 2048w\" sizes=\"auto, (max-width: 2528px) 100vw, 2528px\" \/><figcaption>Chatterbox Multilingual test set: one reference voice, six target outputs.<\/figcaption><\/figure>\n<nav aria-label=\"Table of contents\">\n<p><strong>On this page<\/strong><\/p>\n<ul>\n<li><a href=\"#test\">What the test checked<\/a><\/li>\n<li><a href=\"#outputs\">What each output shows<\/a><\/li>\n<li><a href=\"#speed\">Run time and cost<\/a><\/li>\n<li><a href=\"#choose\">When to choose this model<\/a><\/li>\n<\/ul>\n<\/nav>\n<h2 id=\"test\">What Chatterbox Multilingual voice cloning was tested for<\/h2>\n<p><a href=\"https:\/\/wiro.ai\/models\/resemble-ai\/chatterbox-multilingual\">Chatterbox Multilingual on Wiro<\/a> accepts optional reference audio, text, a language selector, and four generation controls. The docs list 23 language choices: Arabic, Danish, German, Greek, English, Spanish, Finnish, French, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Dutch, Norwegian, Polish, Portuguese, Russian, Swedish, Swahili, Turkish, and Chinese. The same reference clip was reused for all six outputs, so changes come from the target text and settings rather than a new speaker source.<\/p>\n<p>The test probes three practical questions. First, can a neutral English service line stay controlled at the default-style settings? Second, what changes when the script asks for more dramatic delivery? Third, does setting <code>cfg_weight=0<\/code>, the documented language-transfer setting, provide a sensible baseline for Turkish, Japanese, and Spanish with the same cloned voice?<\/p>\n<p><code>exaggeration<\/code> controls expressiveness from 0.25 to 2.0; 0.5 is the documented neutral value and high values can become unstable. <code>cfg_weight<\/code> controls guidance and pace; 0.5 is the balanced recommendation, while 0 is the documented setting for language transfer. <code>temperature<\/code> controls variation. These are short scripts by design: pronunciation, pauses, number wording, and unintended continuation are easier to inspect when the listener has a clear target.<\/p>\n<h2 id=\"outputs\">What the six outputs show<\/h2>\n<h3>1. English support script: the controlled baseline<\/h3>\n<p><strong>Text:<\/strong> Thanks for calling support. Please confirm the last four digits of the order number. Then say the delivery city.<\/p>\n<p><strong>Parameters:<\/strong> language=<code>en<\/code>; exaggeration=<code>0.5<\/code>; cfg_weight=<code>0.5<\/code>; temperature=<code>0.8<\/code>; top_p uses the documented default of <code>1.0<\/code>.<\/p>\n<figure><img loading=\"lazy\" decoding=\"async\" width=\"1200\" height=\"400\" class=\"wp-image-2425\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-1.png\" alt=\"Chatterbox Multilingual voice cloning waveform for English support test\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-1.png 1200w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-1-510x170.webp 510w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-1-900x300.webp 900w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-1-768x256.webp 768w\" sizes=\"auto, (max-width: 1200px) 100vw, 1200px\" \/><figcaption>Output 1: neutral English support script with balanced guidance.<\/figcaption><\/figure>\n<p><audio controls preload=\"none\" style=\"width:100%\"><source src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-test-1-en-support.mp3\" type=\"audio\/mpeg\" \/><\/audio><\/p>\n<p>This is the reference point for the rest of the page. It asks for ordinary transactional speech, including a number-related instruction and a handoff between two sentences. Use this setting first for support calls, product explainers, or any case where steady pacing matters more than performance.<\/p>\n<h3>2. English podcast intro: pacing under denser phrasing<\/h3>\n<p><strong>Text:<\/strong> Welcome back to the show. Today: why latency makes voice apps feel broken. Three quick points: timing, pauses, and turn taking.<\/p>\n<p><strong>Parameters:<\/strong> language=<code>en<\/code>; exaggeration=<code>0.5<\/code>; cfg_weight=<code>0.4<\/code>; temperature=<code>0.8<\/code>; top_p=<code>1.0<\/code>.<\/p>\n<figure><img loading=\"lazy\" decoding=\"async\" width=\"1200\" height=\"400\" class=\"wp-image-2426\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-2.png\" alt=\"Waveform for Chatterbox Multilingual English podcast intro\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-2.png 1200w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-2-510x170.webp 510w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-2-900x300.webp 900w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-2-768x256.webp 768w\" sizes=\"auto, (max-width: 1200px) 100vw, 1200px\" \/><figcaption>Output 2: an English intro with shorter clauses and deliberate punctuation.<\/figcaption><\/figure>\n<p><audio controls preload=\"none\" style=\"width:100%\"><source src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-test-2-en-podcast.mp3\" type=\"audio\/mpeg\" \/><\/audio><\/p>\n<p>The lower guidance value tests whether a slightly looser pace suits a spoken intro. The punctuation and colon make this a useful listen for pause placement. It is not a latency benchmark: the clip is short and the recorded task time includes platform work around synthesis.<\/p>\n<h3>3. English emotional line: controlled expressiveness<\/h3>\n<p><strong>Text:<\/strong> I should have called sooner. The silence made things worse. Please pause. Then say this line slowly: I am sorry.<\/p>\n<p><strong>Parameters:<\/strong> language=<code>en<\/code>; exaggeration=<code>1.2<\/code>; cfg_weight=<code>0.3<\/code>; temperature=<code>0.9<\/code>; top_p=<code>1.0<\/code>.<\/p>\n<figure><img loading=\"lazy\" decoding=\"async\" width=\"1200\" height=\"400\" class=\"wp-image-2427\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-3.png\" alt=\"Waveform for Chatterbox Multilingual expressive English test\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-3.png 1200w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-3-510x170.webp 510w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-3-900x300.webp 900w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-3-768x256.webp 768w\" sizes=\"auto, (max-width: 1200px) 100vw, 1200px\" \/><figcaption>Output 3: an English script designed to reveal emphasis and pause handling.<\/figcaption><\/figure>\n<p><audio controls preload=\"none\" style=\"width:100%\"><source src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-test-3-en-emotion.mp3\" type=\"audio\/mpeg\" \/><\/audio><\/p>\n<p>This is the stress test for the model&#8217;s signature control. Exaggeration rises from 0.5 to 1.2, guidance drops to 0.3, and temperature rises to 0.9. That combination is appropriate when a read needs more character, but it also deserves a human listen before publication. The model documentation warns that extreme exaggeration can become unstable.<\/p>\n<h3>4. Turkish: cross-language transfer baseline<\/h3>\n<p><strong>Text:<\/strong> Merhaba. Bu test ayni sesi Turkce konusturuyor. Lutfen siparis numarasinin son dort hanesini soyle. Sonra teslimat sehrini soyle.<\/p>\n<p><strong>Parameters:<\/strong> language=<code>tr<\/code>; exaggeration=<code>0.6<\/code>; cfg_weight=<code>0<\/code>; temperature=<code>0.8<\/code>; top_p=<code>1.0<\/code>.<\/p>\n<figure><img loading=\"lazy\" decoding=\"async\" width=\"1200\" height=\"400\" class=\"wp-image-2428\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-4.png\" alt=\"Waveform for Chatterbox Multilingual Turkish voice transfer test\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-4.png 1200w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-4-510x170.webp 510w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-4-900x300.webp 900w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-4-768x256.webp 768w\" sizes=\"auto, (max-width: 1200px) 100vw, 1200px\" \/><figcaption>Output 4: Turkish target text using the documented transfer setting.<\/figcaption><\/figure>\n<p><audio controls preload=\"none\" style=\"width:100%\"><source src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-test-4-tr.mp3\" type=\"audio\/mpeg\" \/><\/audio><\/p>\n<p>For a different target language, <code>cfg_weight=0<\/code> is the meaningful variable here. The model docs call it the language-transfer setting. Listen for whether the cloned speaker identity remains useful without carrying the source accent too strongly into Turkish. The script includes a four-digit-order instruction to make pronunciation worth checking rather than relying on a greeting alone.<\/p>\n<h3>5. Japanese: a different writing system and rhythm<\/h3>\n<p><strong>Text:<\/strong> \u3053\u3093\u306b\u3061\u306f. \u3053\u306e\u30c6\u30b9\u30c8\u306f\u540c\u3058\u58f0\u3067\u65e5\u672c\u8a9e\u3092\u8a71\u3057\u307e\u3059. \u6ce8\u6587\u756a\u53f7\u306e\u4e0b4\u3051\u305f\u3092\u8a00\u3063\u3066\u304f\u3060\u3055\u3044. \u305d\u308c\u304b\u3089\u914d\u9054\u5148\u306e\u90fd\u5e02\u3092\u8a00\u3063\u3066\u304f\u3060\u3055\u3044.<\/p>\n<p><strong>Parameters:<\/strong> language=<code>ja<\/code>; exaggeration=<code>0.55<\/code>; cfg_weight=<code>0<\/code>; temperature=<code>0.8<\/code>; top_p=<code>1.0<\/code>.<\/p>\n<figure><img loading=\"lazy\" decoding=\"async\" width=\"1200\" height=\"400\" class=\"wp-image-2429\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-5.png\" alt=\"Waveform for Chatterbox Multilingual Japanese voice transfer test\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-5.png 1200w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-5-510x170.webp 510w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-5-900x300.webp 900w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-5-768x256.webp 768w\" sizes=\"auto, (max-width: 1200px) 100vw, 1200px\" \/><figcaption>Output 5: Japanese target text with zero CFG weight for transfer.<\/figcaption><\/figure>\n<p><audio controls preload=\"none\" style=\"width:100%\"><source src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-test-5-ja.mp3\" type=\"audio\/mpeg\" \/><\/audio><\/p>\n<p>Japanese adds a stronger test of text handling and language rhythm than the Latin-script tests. The parameters remain close to neutral, with only a small lift in exaggeration. This output is a better fit for localization review than for judging raw emotional range: check phrasing, number reading, and whether a native speaker accepts the cadence.<\/p>\n<h3>6. Spanish: concise customer-service localization<\/h3>\n<p><strong>Text:<\/strong> Hola. Este test usa la misma voz en espanol. Di los ultimos cuatro digitos del pedido. Luego di la ciudad de entrega.<\/p>\n<p><strong>Parameters:<\/strong> language=<code>es<\/code>; exaggeration=<code>0.55<\/code>; cfg_weight=<code>0<\/code>; temperature=<code>0.8<\/code>; top_p=<code>1.0<\/code>.<\/p>\n<figure><img loading=\"lazy\" decoding=\"async\" width=\"1200\" height=\"400\" class=\"wp-image-2430\" src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-6.png\" alt=\"Waveform for Chatterbox Multilingual Spanish voice transfer test\" srcset=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-6.png 1200w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-6-510x170.webp 510w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-6-900x300.webp 900w, https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-wave-6-768x256.webp 768w\" sizes=\"auto, (max-width: 1200px) 100vw, 1200px\" \/><figcaption>Output 6: Spanish customer-service text using the same reference voice.<\/figcaption><\/figure>\n<p><audio controls preload=\"none\" style=\"width:100%\"><source src=\"https:\/\/wiro.ai\/blog\/wp-content\/uploads\/2026\/04\/chatterbox-multilingual-test-6-es.mp3\" type=\"audio\/mpeg\" \/><\/audio><\/p>\n<p>This final output repeats the same operational intent in Spanish. It gives a direct check on cross-language reuse for a common support pattern. It should not be treated as proof that every locale needs no review. Brand names, addresses, product vocabulary, and regional variants still need a native-speaker pass.<\/p>\n<h2 id=\"speed\">Run time and cost per output on Wiro<\/h2>\n<table>\n<thead>\n<tr>\n<th>Output<\/th>\n<th>Recorded task elapsed time<\/th>\n<th>Cost shown in available run data<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>English support<\/td>\n<td>44 seconds<\/td>\n<td>Not shown<\/td>\n<\/tr>\n<tr>\n<td>English podcast<\/td>\n<td>10 seconds<\/td>\n<td>Not shown<\/td>\n<\/tr>\n<tr>\n<td>English emotion<\/td>\n<td>7 seconds<\/td>\n<td>Not shown<\/td>\n<\/tr>\n<tr>\n<td>Turkish transfer<\/td>\n<td>10 seconds<\/td>\n<td>Not shown<\/td>\n<\/tr>\n<tr>\n<td>Japanese transfer<\/td>\n<td>9 seconds<\/td>\n<td>Not shown<\/td>\n<\/tr>\n<tr>\n<td>Spanish transfer<\/td>\n<td>10 seconds<\/td>\n<td>Not shown<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>The available records expose elapsed task time for these six runs, but neither the Wiro model documentation nor the records used for this post show a per-output price. No cost has been inferred from another task, model, or platform. The first run is also an outlier, so this small set should not be read as a guaranteed latency figure.<\/p>\n<h2 id=\"choose\">When to pick Chatterbox Multilingual<\/h2>\n<p>Pick Chatterbox Multilingual when one reference voice must travel across multiple supported languages and the project can include listening review. Start near exaggeration 0.5 and cfg_weight 0.5 for ordinary English. For a more expressive read, raise exaggeration carefully and lower guidance slightly. For a new target language with the same reference voice, start at cfg_weight 0, then adjust only after reviewing native pronunciation and accent transfer.<\/p>\n<p>Choose a language-specific model instead when a particular locale, dialect, or specialized vocabulary matters more than one-voice coverage across markets. Choose a low-latency English-only model when real-time turn taking is the priority. For broader agent workflow context, see <a href=\"https:\/\/wiro.ai\/blog\/realtime-voice-conversation-wiro-2026\/\">Realtime Voice Conversation: 3 Smart Wiro Setups in 2026<\/a>, <a href=\"https:\/\/wiro.ai\/blog\/realtime-speech-to-text-wiro-2026\/\">Realtime Speech to Text: 3 Smart Wiro Models in 2026<\/a>, and <a href=\"https:\/\/wiro.ai\/blog\/ai-receptionist-for-missed-calls\/\">AI Receptionist for Missed Calls: 6 Smart Intake Wins<\/a>.<\/p>\n<h2>Source notes<\/h2>\n<p>For model-family details and supported deployment options, see <a href=\"https:\/\/www.resemble.ai\/learn\/models\/chatterbox\" target=\"_blank\" rel=\"noopener\">Resemble AI&#8217;s Chatterbox page<\/a>, the <a href=\"https:\/\/github.com\/resemble-ai\/chatterbox\" target=\"_blank\" rel=\"noopener\">official Chatterbox GitHub repository<\/a>, and the <a href=\"https:\/\/huggingface.co\/ResembleAI\/chatterbox\" target=\"_blank\" rel=\"noopener\">ResembleAI Chatterbox model card on Hugging Face<\/a>. All three sources were reachable when this update was prepared.<\/p>\n<h2>Try it<\/h2>\n<p><a href=\"https:\/\/wiro.ai\/models\/resemble-ai\/chatterbox-multilingual\">Run Chatterbox Multilingual on Wiro<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Chatterbox Multilingual voice cloning is useful only if a reference voice still feels usable when the script, pacing, and language change. This&hellip;<\/p>\n","protected":false},"author":4,"featured_media":2418,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[52],"tags":[166,95,207,62,68],"class_list":["post-2431","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-model-reviews","tag-chatterbox-multilingual","tag-multilingual","tag-speech-to-speech","tag-text-to-speech","tag-voice-clone"],"_links":{"self":[{"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/posts\/2431","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/comments?post=2431"}],"version-history":[{"count":4,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/posts\/2431\/revisions"}],"predecessor-version":[{"id":4271,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/posts\/2431\/revisions\/4271"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/media\/2418"}],"wp:attachment":[{"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/media?parent=2431"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/categories?post=2431"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/wiro.ai\/blog\/wp-json\/wp\/v2\/tags?post=2431"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}