Speechace
  • Introduction
    • Overview
    • Use-Cases
  • Getting Started
    • Pre-requisites
      • API Features
      • Getting the API Key
      • API Regions and endpoints
      • API Limits
    • API Samples
    • Supported Languages
    • API Versioning
    • Authentication
    • Try the Speechace API
    • Error Handling
      • Common Errors
      • Retry Strategies
  • Solutions
    • Speaking Practice for Language Learning
    • Automated Language Assessment with AI
    • Voice AI for Early Literacy
    • Test Prep for Standardized tests
      • PTE Speaking Questions
      • IELTS Speaking Questions
      • TOEFL Speaking Questions
      • CEFR Speaking Questions
      • TOEIC Speaking Questions
    • Speaking Practice in Spanish and French
  • Features
    • Introduction
    • Scripted activities
      • Pronunciation Scoring
        • Word and Sentence pronunciation
        • Multiple choice
        • Custom pronunciations
        • Phoneme list
      • Fluency scoring
        • Passage scoring
      • Lexical stress and intonation
    • Spontaneous activities
      • Open-ended scoring
        • Language scoring
        • Relevance scoring
        • Language detection
      • Task achievement scoring
        • Describe Image
        • Re-tell Lecture
        • Answer Question
  • API Reference
    • Postman API reference
    • Score Text/Pronunciation
      • Handling overall scores
      • Handling word scores
      • Handling phoneme and syllable scores
    • Score Text/Multiple choice
      • Handling multiple choice response
    • Score Text/Markup Language
      • Handling Markup Response
    • Score Text/Stress & Intonation
      • Handing stress and intonation response
    • Score Text/Phoneme list
      • Handling phoneme list response
    • Score Text/Fluency
      • Handling fluency response
      • Fidelity detection
    • Score Text/Validate Text
    • Score Speech/Open-ended
      • Handling language scores
      • Per metric feedback
        • Grammar metrics
        • Vocabulary metrics
        • Coherence metrics
    • Score Speech/Relevance
      • Handling relevance response
    • Score Speech/Language Detection
    • Score Task/Task Achievement
  • Guides on common topics
    • Intepreting quality score
    • Interpreting overall scores
      • Pronunciation Bands
      • Fluency Bands
      • Vocabulary Bands
      • Grammar Bands
      • Coherence Bands
    • Scoring rubrics
    • Interpreting fidelity class
    • Phonetic notation
      • US English (en-us)
      • UK English (en-gb)
      • French (fr-fr, fr-ca)
      • Spanish (es-es, es-mx)
      • US English (en-us)
    • Getting word timestamps in audio
    • Automatic handling of unknown words
    • Phoneme to letter mapping
    • Markup Language
  • Other Resources
    • Requesting Support
    • Rate Limiting
    • Data Retention
    • FAQs
    • Appendices
Powered by GitBook
On this page
  1. Features
  2. Scripted activities
  3. Pronunciation Scoring

Custom pronunciations

PreviousMultiple choiceNextPhoneme list

Last updated 7 months ago

Use for custom pronunciation scoring.

While measuring the pronunciation score, the Speechace engine automatically breaks down the read-aloud target script in to syllables and phonemes using a large language lexicon. However from time to time, developers run in to situations wherein the syllable/phoneme breakup provided by the Speechace API is ambiguous and to resolve such situations, the Speechace API includes a feature that allows developers to prescribe the exact syllable and phoneme breakup they desire. This feature is particularly useful in the following teaching scenarios and can be built by leveraging the function:

  1. Teaching pronunciation of special acronyms, numbers, or terms - Consider the below demo wherein user is asked to say: In this example, the user can say "Agent 0 0 7 worked for M I 6" or the user can say "Agent Double O 7 worked for M I 6", the latter being the more stylish way of speaking. Here a developer can use the Score Text Markup Language function to prescribe the desired pronunciation of the word 007 as Double O 7 to the Speechace API and then the Speechace API will only give a good pronunciation score if the user says Double O 7. If the user instead says zero, zero, 7 then they will not get a good score from the Speechace API.

Teaching correct pronunciation of Heteronyms - In some languages such as English, there are words called heteronyms that are spelled alike but they have different meanings. As an example, consider the word "Read" which can be pronounced as reed or red. If you use the Speechace API to score the pronunciation of the string Read then by default it will accept both reed and rehd and return syllable and phoneme breakup for the best matching pronunciation. This may not be ideal as only one of the pronunciations may actually be correct given the larger sentence. In such cases, developers can pass a hint to the Speechace API to suggest the correct pro You can experience this mechanism in the below demo: You can try saying "He reed his fragments aloud" and you should see the results in the below image. As can be observed, the word "read" is marked in Red and the system suggests that the "EH" sound is missing. If try to pronounce the word read as 'rehd' and the system will mark the word "read" in Green.

Teaching correct syllables and phonemes - As mentioned in the above description, there are a few rare times when the Speechace API may not break up the string in to the best combination of syllables and phonemes. Under such circumstances, developers can specify the desired combination of syllables and phonemes using the function. As an example, in the following demo, the Speechace API may return the word "nothing" as a single syllable, which is incorrect. Therefore a developer may provide a hint to the Speechace API that the word "nothing" is componsed of two syllables: noth & ing:

Say: He read his fragments aloud.
Score Text/Markup Language
Score Text/Markup Language
Score Text/Markup Language
Agent Double O 7 worked for M I 6