mixed text IPA

IPA (International Phonetic Alphabet)

What is IPA?

IPA is the global standard for foreign language pronunciation: the International Phonetic Alphabet. IPA provides a consistent representation of sounds across multiple languages. Essentially a notation system capturing the segments of sounds that make up words (phonemes), it was devised in the 19th century by French and British language teachers, led initially by Paul Passy. IPA is used in many print and online dictionaries to aid pronunciation. The International Phonetic Alphabet is maintained by the International Phonetic Association. Today, many conservatoires, music schools, singing teachers, and vocal coaches use IPA to train singers in high-quality language diction in sung performance. Singers often use IPA symbols to annotate scores in preparation for performance.

Why an IPA tool for song?

When browsing a song in our Discover Song database, the IPA button is designed to support singers’ diction when performing in a foreign language. Singers work repeatedly across multiple languages and need reliable, song-specific pronunciation guides to help prepare for performance. Sung French, for example, is different from spoken French in terms of how to pronounce the /r/ consonant, whether to pronounce the /e/ vowel that is normally silent at the end of many words, or what to do with liaisons between words. Typical training approaches or preparation tools that exist can be very prescriptive, and sometimes also very subjective, without clear rationale for certain decisions. Other resources tend to provide IPA guides only for specific songs, e.g. in Handbooks for Singers, or editions of songs with prefatory guides, but these are not comprehensive and singers may not have access to the necessary resource for the songs they are working on. Our IPA tool can cover the whole gamut of songs (including new ones being introduced to the repertoire). The tool also recognises that pronunciation habits change over time, as languages also change, so it includes options and variants. Different regional accents and historical versions of a language also impact what is suitable or appropriate in terms of diction.

Oxford Song's Discover Song resource includes a c.4000-song database spanning more than 20 languages. We have prioritised the two major languages of song—French and German—and hope to extend to other languages in the future. We have also created a design which allows any song to be processed and a high-quality IPA transcription produced.

How to use the IPA tool?

When browsing songs via our database, click the IPA button to see the IPA transcription of a poem alongside the original language text presented line-by-line. Where options are available, these are highlighted, and a click takes you to short guides explaining why the options exist. Where a word is not in the IPA dictionary (e.g. a rare term, or a loan word from another language, or proper noun such as place names or names of figures from Greek mythology or characters from commedia dell’arte), these are flagged as possible IPA transcriptions— use particular judgement and care, consulting other sources where these are available, because while helpful, these are derived from a probabilistic tool, not a confirmed IPA dictionary term. Next to each line of text/IPA, there is a ‘playback’ button which you can click to hear an audio example. Note that for technical reasons the audio examples may not currently include all the song-specific pronunciation elements, but they offer a useful guide. There are options to hear male/female voice and different playback speeds provided at the top of the page.

An IPA table guide offering examples is also provided for guidance, to help you if you are unsure of how to pronounce an IPA symbol.

While the IPA function has been developed to integrate complex poetic features, such as the liaisons and the mute ‘e’ in French, singers might want to use it in combination with the music score to cross check additional information given by the composer. For example, stresses on syllables are not always shown in the IPA transcription but they might be shown through musical notation.

A note on the ‘r’ sound in French and German: in song ‘r’ is traditionally pronounced with a flip of the tongue, which is represented with the /r/ IPA symbol. However, more and more singers prefer to follow the spoken usage, such as a more guttural ‘r’ in sung French, or regional variants in everyday German, for example. Our IPA function uses the /r/ symbol, but singers are free to follow the spoken usage.

How does the IPA tool work?

The technology uses IPA transcriptions from large-scale open-source IPA dictionaries, and supplements these with a set of rules which implement aspects of pronunciation relevant to poetry, versification, and prosody within the context of song. As far as we know, this is the only IPA tool able to implement versification rules and song conventions.

To ensure the best quality IPA transcriptions, in which singers can have confidence, we have also added new morphological variants to the lemma (dictionary entry), such as plural endings.

To account for words not already in the open-source IPA dictionaries, the IPA tool integrates an open-source probabilistic tool called Epitran. These Epitran transcriptions of unusual words mean that there are no gaps in the IPA transcription, although words transcribed through the Epitran system should be treated with greater care and singers should use their judgment and knowledge (they are flagged where appropriate).

The system has been designed with consideration of how to handle:

  • Old language, e.g. old French which uses different spellings and for which a standardised modern pronunciation may be inappropriate.
  • Loan words, or words in another language used in a song, e.g. ‘Piccadilly’ or ‘gentleman’ used in Le diva de l’Empire by Éric Satie (lyrics by Dominique Bonnaud & Numa Blès).

although IPA for these kinds of texts/words continue to be refined as we use the tool.

Limitations include:

  • How successfully it can handle contracted words (e.g ‘wart’ ich’ in German).
  • Primary and secondary syllable stress are not always shown (but are usually indicated in the vocal writing in a song score).

How was the IPA tool developed?

The IPA tool was developed as a collaboration between Professor Helen Abbott and Oxford Song, extending the existing artistic partnership between Oxford Song and the University of Birmingham.

Project lead: Helen Abbott (University of Birmingham)
Developers: Steven Abbott (TCNF), Sean Cooper (Order & Chaos Creative)
Web implementation: Tom King (OxAlto)
Project manager: Nina Rolland (University of Birmingham / Northeastern University London)

Language checkers: Harvey Jones (formerly University of Birmingham), Claire Ward (soprano and former Oxford Song Young Artist Platform Finalist), Héloïse Bernard (soprano and former Oxford Song Mastercourse Artist)

Large-scale open-source phonetic dictionaries were used as the starting point for the tool and supplemented by other resources such as Epitran. Human intervention was used throughout to sense-check and test accuracy of the prototype initially developed in 2021.

The tool was built in JavaScript, and runs on AWS Lambda.

The development of the tool was made possible through the support of generous philanthropic funding from a donor who wishes to remain anonymous.

© Helen Abbott & Nina Rolland 2023

Select Tickets