Back to Blog
Languages

German Text to Speech: Compound Words and Natural Voices

German text to speech must cope with long compounds and three standard varieties. See which voices exist, how to fix pronunciation and how to start free.

9 min read
By OndaLabs Team

German text to speech turns German text into spoken audio with a synthetic voice. It has to handle very long compound words, umlauts, the ß and three standard varieties: German, Austrian and Swiss Standard German. The major providers now list German voices, so you can pick one, fix hard words and generate a passage.

This guide explains why compounds trip up synthetic voices, how to fix a mispronounced word, which German voices the main providers list as of October 2026, and how to make German audio with OndaLabs.

Why German compound words are hard for text to speech

German writes most compounds as one word. Haustür ("house door") is a single word, not two, and the same rule covers long nouns built from many parts. Compounds that include letters, abbreviations or numbers are hyphenated instead, as in A-Dur.

The table below lists well-known long words from the sources. The words use the ae, oe and ue spellings those sources give, so the letter counts match.

| Word | Letters | What the source says | |---|---|---| | Rechtsschutzversicherungsgesellschaften | 39 | Called "the longest word in regular use", attributed to the Guinness Book of Records | | Aufmerksamkeitsdefizit-Hyperaktivitaetsstoerung | 42 | Described as the longest word in the Duden | | Rindfleischetikettierungsueberwachungsaufgabenuebertragungsgesetz | 63 | Removed from the books in 2013 after EU rules made the law obsolete | | Donaudampfschiffahrtselektrizitaetenhauptbetriebswerkbauunterbeamtengesellschaft | 79 | Named the longest published German word by the 1972 Guinness Book of World Records |

Sources: Longest word in German and the Wikipedia article on German orthography.

Numbers can be compounds too. German whole numbers under one million can be written as a single word. 777,777 becomes siebenhundertsiebenundsiebzigtausendsiebenhundertsiebenundsiebzig, which is 65 letters.

Long words are rare in running text. Newly coined or unusual compounds are the place to check first, because a misreading is most likely there. This is general guidance. This research found no benchmark that compares how engines read compounds.

How to fix a mispronounced German word

Start with the cheapest fix. Respell the word in your script, for example by hyphenating it so the syllables are clearer. This works in any tool. If it is not enough, the sources document three routes:

  • Azure phoneme tags. Azure's phoneme control for German uses the sapi phone set in de-DE, de-AT and de-CH. Syllables are separated by hyphens, and a digit after the stressed vowel marks stress: 1 for primary, 2 for secondary. Azure's example for Allgemeinwissen is a 2 l - g ax - m ay 1 n - v ih - s n. A glottal stop (^) is needed between two distinct vowels (Azure phonetic sets).
  • Azure custom lexicons. A lexicon is a W3C PLS file of up to 100 KB, with one locale per file. Matching is case sensitive, and a change can take up to 15 minutes to apply because of caching (Azure pronunciation markup).
  • Google Chirp 3 HD custom pronunciation. Chirp 3 HD accepts IPA or X-SAMPA through the SSML phoneme tag. Its SSML support is partial and in Preview, and it is not available for streaming (Google Cloud Chirp 3 HD).

German voices from the main providers as of October 2026

Voice counts change often, so the figures below are dated. Counts marked approximate come from a summary of the provider's list.

| Provider | German voices listed | Pronunciation control documented here | |---|---|---| | Azure | de-DE: 23 entries, including HD voices Seraphina and Florian and standard voices such as Katja and Conrad. de-AT: Ingrid and Jonas. de-CH: Leni and Jan. | SSML phoneme (sapi) and custom lexicon | | Amazon Polly | de-DE: Vicki, Daniel, Lennart, Marlene and Hans. de-AT: Hannah. de-CH: Sabrina. | Not checked in this research | | Google Cloud | Chirp 3 HD supports de-DE. Roughly 21 de-DE voices across several types (approximate). | IPA or X-SAMPA through the SSML phoneme tag (Preview) | | ElevenLabs | German on Multilingual v2, Flash v2.5, v3 and v4 | Number normalisation settings |

Azure's counts include HD and Flash variants. Polly's voices sit on different engines. Vicki offers generative, neural and standard. Daniel, Hannah and Sabrina offer generative and neural. Lennart is generative only, and Marlene and Hans are standard only.

How German accents and spelling differ

The sources name three standardised varieties: German Standard German, Austrian Standard German and Swiss Standard German. The de-AT and de-CH locale labels map to the Austrian and Swiss voices above. A locale label does not prove that a voice speaks a regional dialect. The sources do not say that any of these voices speaks Swiss German dialect, so treat a de-CH voice as Standard German under a Swiss label.

Spelling follows the variety too. ß is used only after long vowels and diphthongs, and it is not used in Switzerland or Liechtenstein. For Swiss readers, write ss. Capital ẞ became an official letter on 29 June 2017.

Pronunciation varies inside the language as well. Coda devoicing occurs in northern varieties but not in southern Germany, Austria or Switzerland. The glottal stop is much more frequent in northern varieties.

How German numbers and abbreviations are read

Azure supports say-as types for German locales, including cardinal, ordinal, date, time, currency, telephone and unit. Characters and spell-out work in all locales.

ElevenLabs' Flash v2.5 does not normalise numbers by default, to keep latency low. That affects phone numbers, dates and currencies. Enterprise customers can switch normalisation on with the apply_text_normalization option. ElevenLabs recommends Multilingual v2 when number reading matters, or normalising the text before you send it.

If a number must be read exactly, write it out in the script, then check the result by ear.

Writing a German script that sounds natural

  • Pick the variety first, then use ß or ss to match your audience and keep the spelling consistent.
  • Write compounds in standard spelling. Fix a misreading with the methods above rather than splitting the word.
  • Spell out abbreviations the first time they appear, unless the listener should hear the abbreviation.
  • Break long passages into paragraphs, generate each one, and listen to the joins.
  • Test the hardest sentences first. Compounds, numbers and names show problems fastest.

Making German audio with OndaLabs

  1. Open the text to speech studio and paste your German script.
  2. Pick one of the 30 prebuilt voices, which work in every supported language. German has its own page: German text to speech on OndaLabs.
  3. Choose a model. Onda Lite is available on all plans at 1 credit per character. Onda Flash, on paid plans, is the most expressive and costs 1.5 credits per character.
  4. Describe the delivery in plain words, or add a supported expressive tag where it helps. Generate a short test first.
  5. Respell any word that comes out wrong, then regenerate.

OndaLabs documents plain-words direction and expressive tags. It does not document phoneme tags, so respell hard words in the script instead.

A 1,000-character German passage costs 1,000 credits on Onda Lite and 1,500 on Onda Flash. The Free plan is $0 and includes 2,000 credits a month, about 2 minutes of audio. Starter is $9 a month and includes 170,000 credits, about 189 minutes, Onda Flash and commercial use. Full plan details are on the pricing page.

If you are also producing Portuguese, read Portuguese text to speech: Brazil vs Portugal for the variant choice. For a whole video in another language, see AI dubbing explained.

Frequently asked questions

Is there a free German text to speech tool?

OndaLabs has a Free plan at $0 with 2,000 credits a month, about 2 minutes of audio on Onda Lite. That is enough to test a German script. Check the current terms of any other tool before you rely on its free plan.

Which German AI voice sounds most natural?

This guide cannot rank voices. No benchmark in the research compared German voices, and the provider pages list voices without rating them. Generate the same paragraph with two or three voices and listen to each.

Can I use a German AI voice commercially?

OndaLabs includes commercial use on Starter and above, from $9 a month. Check each provider's terms before you publish. OpenAI's usage policy, for example, requires clear disclosure to end users that a TTS voice is AI-generated.

Can text to speech do Austrian or Swiss German accents?

Azure and Amazon Polly list voices under de-AT and de-CH locales. The sources do not show that these voices speak regional dialects, so treat them as Standard German under those labels. Listen to a sample before you choose one.

How do I fix a mispronounced German word?

Respell or hyphenate the word first. For exact control, Azure supports German phoneme tags and custom lexicons, and Google Chirp 3 HD accepts IPA or X-SAMPA in Preview. OndaLabs documents no phoneme tags, so respelling is the route to use there.

Does German text to speech read numbers and dates correctly?

Azure supports say-as types for German, such as date, time and currency. ElevenLabs Flash v2.5 does not normalise numbers by default. Test any number that must be read exactly, and write it out if the reading matters.