ElevenLabs SSML Support: Model Compatibility, Limitations, and Audio Tags Alternative

SSML Support by Model

SSML (Speech Synthesis Markup Language) is a W3C-based XML standard used to control pauses, pronunciation, and pacing in TTS systems. ElevenLabs' support varies significantly by model version.

• Common Support: The <break> tag is supported on all models except Eleven v3 and Eleven v4, allowing pauses up to 3 seconds.
• Pronunciation Control: The <phoneme> tag for custom pronunciation (using IPA or CMU Arpabet) is supported only on Eleven Flash v2 and Eleven Turbo v2, and is restricted to English.
• No SSML Support: Eleven v3 and Eleven v4 do not support standard SSML tags. Instead, they use Audio Tags and native IPA support.

WebSocket API Configuration

SSML parsing is disabled by default in the WebSocket API. Developers must set the enable_ssml_parsing parameter to true to activate it. For optimal performance, each SSML tag should be sent as a single, complete message to avoid latency caused by the parser waiting for split chunks. Note that uploading a pronunciation dictionary automatically enables SSML parsing regardless of this setting.

Unsupported Tags and Alternatives

Many standard SSML tags are not supported by ElevenLabs, requiring alternative methods:

• <prosody> (pitch, speed, volume): Use Voice settings or Audio Tags like [slowly] or [rushed] in v3/v4.
• <emphasis>: Use capitalization, punctuation, or Audio Tags like [shouts].
• <say-as> (numbers, dates): Use text normalization or write out numbers in the script.
• <lang>: Use multilingual models (e.g., Multilingual v2, Flash v2.5) and write text in the target language.
• <voice>: Generate separate audio for each speaker or use the multi-speaker dialogue feature in v3/v4.

Audio Tags as the Modern Alternative

Audio Tags are a natural language control system introduced in Eleven v3 and Eleven v4 to replace rigid SSML syntax. They allow users to insert instructions directly into the text using brackets, such as [whispering], [sighs], [applause], or [French accent].

• Contextual Awareness: Unlike SSML, which applies instructions at fixed points, Audio Tags interpret cues based on surrounding context, resulting in more natural prosody and tone.
• Native IPA Support: Eleven v4 supports inline IPA pronunciation using slashes (e.g., /ˈfɒnɪks/) without XML tags, covering over 90 languages.
• Workflow Integration: Audio Tags enable control over emotion, speed, and sound effects within the ElevenCreative workflow, simplifying the creation of final audio and video outputs.