Selected SSML Source Auditor

How to use Selected SSML Source Auditor

Inspect author-supplied speech markup before sending it to a synthesis system. The evidence keeps written words, substitution aliases, opaque IPA strings, processor-dependent say-as controls and explicit break times distinct. A clearly labeled projection adds structure/control markers and never claims to reproduce actual spoken output.

1. Paste a supported SSML 1.1 source document containing only the declared small element/attribute profile.
2. Run local review and inspect written text, aliases, opaque pronunciation controls and exact explicit pauses separately.
3. Download evidence and verify the intended result in your own synthesis system only after reviewing privacy and provider-specific behavior.

Good to know

• At most 50,000 source characters, 500 elements, 20 levels and 1,000 source segments. Native XML parsing rejects malformed XML, DTD/entities and processing instructions. Root speak requires the SSML namespace, version 1.1 and a nonblank xml:lang source value; language correctness is not inferred.
• This subset supports only speak, p directly under speak, s under speak/p, text-only sub/phoneme/say-as and empty break/mark. Attributes are explicitly allowed per element; phoneme requires declared ipa, whose phonetic correctness is unassessed. Voice/prosody/emphasis/lang/token/metadata, audio/lexicon/lookup, XML base, foreign namespaces, unknown attributes and provider extensions are refused.
• Break times require nonnegative plain numbers with up to three decimals, leading digits and lowercase ms or s. Each break is at most 5 seconds; total explicit breaks are at most 300 seconds. Exact microsecond accounting does not include unspecified breaks, normal speech, prosody or processor-added pauses and is not an audio duration estimate.
• Projection applies only declared sub aliases, keeps written phoneme/say-as text and inserts labeled boundaries/controls. Language, interpretation and pronunciation remain synthesis-dependent. Original source and aliases can be private; no resource/schema is fetched and no markup is mounted, executed or synthesized. Real native XML positives remain a release gate.
• Processing is local. Output is limited to 1,000,000 text characters unless a stricter file limit is stated.