Phoneme guide: Chinese (cjk group)¶
Every phoneme token of mbfa-cjk (48 tokens: 44 phonemes, 3 special tokens and the model's internal blank). The web interface shows the same text when you hover a phoneme.
How to read the tables:
- Sounds like is an approximation. The example words are the real thing: they are the most frequent words containing the token in the FLEURS test sentences (dev split) of 2 languages (Mandarin, Cantonese), written in the G2P's own tokens, which is exactly what the aligner sees.
- Used most in gives the languages that use the token most, as a share of all the phonemes of that language.
- Also used for: IPA the G2P wrote that has no token of its own in this model, so it was mapped to this one.
- Rare tokens never appear in those sentences: they are very rare in this language.
Tone is a separate layer, not part of these tokens: see the tone guide. Affricates and diphthongs are split into tokens: 在 = t s a ɪ, 这 = ʈ ʂ ɤ; an aspirated affricate puts the aspiration on its second half (其 = t ɕʰ i, 此 = t sʰ i). Give each part its own span. The aspirated stops pʰ tʰ kʰ are the sounds written p t k in pinyin and Jyutping; the plain p t k are the ones written b d g.
Where boundaries go¶
Suggested conventions. Agree on them before annotating, so that every annotator places boundaries the same way.
- Stops (p t k b d g …). A stop starts where the closure starts: energy drops to a near-silent gap, and the formants of the vowel before it end. It ends after its burst (and after its aspiration, for ʰ tokens), where the next sound begins.
- Vowels. Start at the first regular glottal pulse with clear formants; end where the formants fade or the next consonant's closure or noise begins.
- Fricatives (s ʃ f x …). The span of the noise: start and end where the high-frequency noise starts and stops.
- Nasals (m n ŋ …). A steady, low murmur with weak upper formants. The edges are usually sudden changes in energy.
- Glides, laterals, r (j w l ɹ …). Edges are smooth formant transitions: put the boundary at the middle of the transition.
- Diphthongs (a͡ɪ …). One token for both parts: from the start of the first vowel to the end of the second. Do not split it.
- Pauses. Leave a short pause between words as a gap, or insert SIL where the speaker clearly stops.
Diacritics¶
Marks added to a base symbol change it as follows.
| mark | name | meaning |
|---|---|---|
| ◌̃ | nasalised | air also flows through the nose (French “an”, Portuguese “não”) |
| ◌̯ | non-syllabic | the weak, glide-like half of a diphthong; it is not a syllable on its own |
| ◌̩ | syllabic | forms a syllable with no vowel, as the n of “button” |
| ◌̥ | voiceless | said without vocal-fold vibration, as if whispered |
| ◌̪ | dental | the tongue tip touches the upper teeth |
| ◌̠ | retracted | made a little further back in the mouth |
| ◌̝ | raised | a little closer (tongue higher) than the plain symbol |
| ◌̞ | lowered | a little more open; a lowered fricative is a smooth approximant with no friction |
| ◌̺ | apical | made with the very tip of the tongue |
| ◌̻ | laminal | made with the blade of the tongue, just behind the tip |
| ◌̆ | extra-short | a very short version of the vowel |
| ◌́ | accent ´ (high) | the G2P's dictionary marks a pitch accent or stress here |
| ◌̀ | accent ` (low) | the G2P's dictionary marks a pitch accent or stress here |
| ◌̂ | accent ˆ (falling) | the G2P's dictionary marks a falling pitch accent |
| ◌̌ | accent ˇ (rising) | the G2P's dictionary marks a rising pitch accent |
| ʲ | palatalised | said with the tongue body raised toward the palate: a y-like colour (“soft” consonants) |
| ʰ | aspirated | followed by a puff of breath before the vowel starts (English “top” has one, “stop” does not) |
| ʷ | labialised | said with rounded lips |
| ʼ | ejective | a pop made by the throat, with no breath from the lungs |
| ʱ | breathy-voiced | voiced, then released with a breathy, murmured puff (Hindi “bh”, “gh”) |
| ˤ | pharyngealised (emphatic) | said with the tongue root pulled back: a darker, “heavy” sound (Arabic ط, ص) |
| ◌̚ | unreleased | a final stop with no audible burst: the closure is held and the sound ends silently |
| ◌̈ | centralised | said closer to the centre of the mouth than the plain vowel |
Special tokens¶
| token | meaning |
|---|---|
SIL |
silence. A pause: no speech. Use it for a pause the speaker makes inside a segment. |
noise |
noise. A non-speech sound: a breath, cough, laugh or click. |
<unk> |
unknown. The G2P wrote a sound this model has no token for. Keep its time span; the label cannot say what it is. |
Vowels¶
| token | name | sounds like | examples | used most in |
|---|---|---|---|---|
| a | open front unrounded vowel | a as in Spanish “casa” | Mandarin 在 /t s a ɪ/; Cantonese 國家 /k w ɔ k k a/ | Mandarin 5.1%, Cantonese 3.8% |
| i | close front unrounded vowel | ee as in “see” | Cantonese 世界 /s ɐ i k a i/; Mandarin 的 /t i/ | Cantonese 9.5%, Mandarin 4.9% |
| e | close-mid front unrounded vowel | e as in Spanish or Italian “de”; the start of English “say” | Mandarin 为 /w e ɪ/; Cantonese 非常 /f e i s œ ŋ/ | Mandarin 2%, Cantonese 1.4% |
| o | close-mid back rounded vowel | o as in Spanish or Italian “no” | Cantonese 都 /t o u/; Mandarin 有 /j o ʊ/ | Cantonese 1.4%, Mandarin 1.4% |
| u | close back rounded vowel | oo as in “food” | Cantonese 有 /j ɐ u/; Mandarin 不 /p u/ | Cantonese 4.8%, Mandarin 3.9% |
| ɪ | near-close front unrounded vowel | i as in “sit” | Mandarin 在 /t s a ɪ/; Cantonese 的 /t ɪ k/ | Mandarin 5.4%, Cantonese 3.8% |
| ə | mid central vowel (schwa) | the weak vowel of “about”, “sofa” | Mandarin 人 /ʐ ə n/; Cantonese ehrenstrahl /ɛ |
Mandarin 1.4% |
| ɛ | open-mid front unrounded vowel | e as in “bed” | Mandarin 年 /n j ɛ n/; Cantonese 這個 /t s ɛ k ɔ/ | Mandarin 1.8%, Cantonese 0.82% |
| ɑ | open back unrounded vowel | a as in “father” | Mandarin 了 /l j ɑ ʊ/; Cantonese ehrenstrahl /ɛ |
Mandarin 3.8% |
| ɔ | open-mid back rounded vowel | o as in British “thought” | Cantonese 個 /k ɔ/; Mandarin 它 /tʰ u ɔ/ | Cantonese 4.1%, Mandarin 1.7% |
| ʊ | near-close back rounded vowel | oo as in “book” | Mandarin 了 /l j ɑ ʊ/; Cantonese 不同 /p ɐ t tʰ ʊ ŋ/ | Mandarin 4.8%, Cantonese 1.8% |
| ɯ | close back unrounded vowel | oo said with spread lips (Turkish “ı”) | Mandarin 斯 /s ɯ/ | Mandarin 0.45% |
| ɐ | near-open central vowel | a short, relaxed “uh” (Portuguese “cama”, German “-er” endings) | Cantonese 有 /j ɐ u/ | Cantonese 6.7% |
| y | close front rounded vowel | ee said with rounded lips (French “tu”, German “über”) | Cantonese 最 /t s ɵ y/; Mandarin 与 /y/ | Cantonese 2.4%, Mandarin 1% |
| ɵ | close-mid central rounded vowel | a rounded vowel between e and o (Swedish short u) | Cantonese 最 /t s ɵ y/ | Cantonese 1.4% |
| œ | open-mid front rounded vowel | e as in “bed” said with rounded lips (French “neuf”, German “können”) | Cantonese 非常 /f e i s œ ŋ/; Mandarin 滑雪 /x w a ɕ ɥ œ/ | Cantonese 1.2%, Mandarin 0.78% |
| ɤ | close-mid back unrounded vowel | o said with spread lips (Estonian “õ”) | Mandarin 和 /x ɤ/ | Mandarin 3.2% |
| ɚ | r-coloured schwa | a weak vowel with an r colour (American “butter”, Mandarin “er”) | Mandarin 是 /ʂ ɚ/ | Mandarin 1.8% |
Glides and non-syllabic vowels¶
| token | name | sounds like | examples | used most in |
|---|---|---|---|---|
| j | palatal approximant | y as in “yes” | Mandarin 了 /l j ɑ ʊ/; Cantonese 有 /j ɐ u/ | Mandarin 4.4%, Cantonese 4.3% |
| w | labial-velar approximant | w as in “we” | Mandarin 为 /w e ɪ/; Cantonese 國家 /k w ɔ k k a/ | Mandarin 2.5%, Cantonese 2.3% |
| ɥ | labial-palatal approximant | y said with rounded lips (French “huit”) | Mandarin 滑雪 /x w a ɕ ɥ œ/ | Mandarin 0.78% |
Stops¶
| token | name | sounds like | examples | used most in |
|---|---|---|---|---|
| t | voiceless alveolar stop | t as in “ten” | Cantonese 的 /t ɪ k/; Mandarin 的 /t i/ | Cantonese 11%, Mandarin 8.8% |
| k | voiceless velar stop | k as in “key” | Cantonese 的 /t ɪ k/; Mandarin 个 /k ɤ/ | Cantonese 7%, Mandarin 1.7% |
| p | voiceless bilabial stop | p as in “pen” | Cantonese 不同 /p ɐ t tʰ ʊ ŋ/; Mandarin 不 /p u/ | Cantonese 2.3%, Mandarin 1.8% |
| ʈ | voiceless retroflex stop | t with the tongue tip curled back (Swedish and Norwegian “rt”) | Mandarin 这 /ʈ ʂ ɤ/ | Mandarin 3.1% |
| tʰ | voiceless alveolar stop, aspirated | t as in “ten”; aspirated: followed by a puff of breath before the vowel starts (English “top” has one, “stop” does not) | Mandarin 它 /tʰ u ɔ/; Cantonese 不同 /p ɐ t tʰ ʊ ŋ/ | Mandarin 1.2%, Cantonese 1.1% |
| kʰ | voiceless velar stop, aspirated | k as in “key”; aspirated: followed by a puff of breath before the vowel starts (English “top” has one, “stop” does not) | Mandarin 克 /kʰ ɤ/; Cantonese 以及 /j i kʰ ɐ p/ | Mandarin 0.9%, Cantonese 0.8% |
| pʰ | voiceless bilabial stop, aspirated | p as in “pen”; aspirated: followed by a puff of breath before the vowel starts (English “top” has one, “stop” does not) | Mandarin 普 /pʰ u/; Cantonese 普遍 /pʰ o u pʰ i n/ | Mandarin 0.38%, Cantonese 0.34% |
Fricatives¶
| token | name | sounds like | examples | used most in |
|---|---|---|---|---|
| s | voiceless alveolar fricative | s | Cantonese 在 /t s ɔ i/; Mandarin 在 /t s a ɪ/ | Cantonese 7.7%, Mandarin 1.8% |
| f | voiceless labiodental fricative | f | Cantonese 非常 /f e i s œ ŋ/; Mandarin 方法 /f ɑ ŋ f a/ | Cantonese 1.2%, Mandarin 0.96% |
| h | voiceless glottal fricative | h as in “hat” | Cantonese 許多 /h ɵ y t ɔ/; Mandarin rockhampton / |
Cantonese 1.8%, Mandarin 0.02% |
| x | voiceless velar fricative | ch as in Scottish “loch”, German “Bach” | Mandarin 和 /x ɤ/ | Mandarin 1.6% |
| ʂ | voiceless retroflex fricative | a hard sh with the tongue tip back (Polish “sz”) | Mandarin 是 /ʂ ɚ/ | Mandarin 4.5% |
| ɕ | voiceless alveolo-palatal fricative | a soft sh, with the tongue body raised (Polish “ś”) | Mandarin 进行 /t ɕ ɪ n ɕ ɪ ŋ/ | Mandarin 3.8% |
| ʐ | voiced retroflex fricative | a hard zh (Polish “ż”, “rz”) | Mandarin 人 /ʐ ə n/ | Mandarin 0.79% |
| sʰ | voiceless alveolar fricative, aspirated | s; aspirated: followed by a puff of breath before the vowel starts (English “top” has one, “stop” does not) | Cantonese 因此 /j ɐ n t sʰ i/; Mandarin 曾 /t sʰ ɤ ŋ/ | Cantonese 1.6%, Mandarin 0.38% |
| ɕʰ | voiceless alveolo-palatal fricative, aspirated | a soft sh, with the tongue body raised (Polish “ś”); aspirated: followed by a puff of breath before the vowel starts (English “top” has one, “stop” does not) | Mandarin 其 /t ɕʰ i/ | Mandarin 1% |
| ʂʰ | voiceless retroflex fricative, aspirated | a hard sh with the tongue tip back (Polish “sz”); aspirated: followed by a puff of breath before the vowel starts (English “top” has one, “stop” does not) | Mandarin 车 /ʈ ʂʰ ɤ/ | Mandarin 0.91% |
Nasals¶
| token | name | sounds like | examples | used most in |
|---|---|---|---|---|
| n | alveolar nasal | n | Mandarin 年 /n j ɛ n/; Cantonese 人 /j ɐ n/ | Mandarin 6.3%, Cantonese 4.9% |
| m | bilabial nasal | m | Cantonese 動物 /t ʊ ŋ m ɐ t/; Mandarin 名 /m ɪ ŋ/ | Cantonese 2.1%, Mandarin 1.2% |
| ŋ | velar nasal | ng as in “sing” | Cantonese 影響 /j ɪ ŋ h œ ŋ/; Mandarin 上 /ʂ ɑ ŋ/ | Cantonese 6.1%, Mandarin 5.7% |
| m̩ | bilabial nasal, syllabic | m; syllabic: forms a syllable with no vowel, as the n of “button” | rare |
Laterals¶
| token | name | sounds like | examples | used most in |
|---|---|---|---|---|
| l | alveolar lateral approximant | l as in “leaf” | Cantonese 能力 /l ɐ ŋ l ɪ k/; Mandarin 了 /l j ɑ ʊ/ | Cantonese 2.1%, Mandarin 1.8% |
Rhotics (r sounds)¶
| token | name | sounds like | examples | used most in |
|---|---|---|---|---|
| ɻ | retroflex approximant | an English-like r with the tongue tip curled back | Mandarin 尔 /a ɻ/ | Mandarin 0.14% |