Voicing · p/b · t/d · k/ɡ The difference between pata and bata is not aspiration — Spanish p and b are both unaspirated. What separates them is whether the vocal folds vibrate. Mandarin speakers replace the voiced stop with an unaspirated voiceless one, which a native ear hears as plain p.
Same position, one thing changed
Each pair below was recorded by the same speaker, with the same vowel, the same stress and the same sentence position. Only the voicing of the initial consonant differs. Play the two cells in alternation and listen for the buzz.
Four steps to find the vibration
The order matters: feel the vibration first, then the word onset, then the sentence.
Touch before you listen
Press a finger to your larynx and say the Mandarin b (as in bà), then p (pà). Neither vibrates — Mandarin b is an unaspirated voiceless stop, not a voiced one. That is the starting point: your native language has no voiced stops.
Put the vibration back
Keep the b articulation (lips closed, then released) but start the vocal folds before releasing. Try a soft m first — that already vibrates — and open the lips into b. It sounds like mb, and that is correct: it is what Spanish speakers do for b after a pause.
Practise the pairs word-initially
Start with onsets only: pata / bata, peso / beso, casa / gasa. Say each pair three times, checking your throat once per pair, then record yourself and compare with the audio here. If your bata sounds the same as pata, the vibration is missing.
Move into sentences
Every audio clip on this page is a full sentence. Sentence position is harder: the same b is a stop after a pause but loosens between vowels — that is spirantisation, covered on another page. Here you only need to keep the folds vibrating.
Which one are you
The three most common readings, each with a different starting point.
bata and pata come out identical
The classic case. You are producing an unaspirated voiceless stop where Spanish needs a voiced one — unavoidable if your native language has no voicing contrast.
My p comes out with a puff of air
An English or pinyin habit. Spanish p is unaspirated; hold your hand in front of your mouth and say pata — you should feel almost nothing.
Fine in isolated words, blurred in sentences
Expected. Voiceless stops can pick up voicing from a preceding voiced sound. Get the onset stable first; connected speech has its own page.
Word list
16 words grouped by where the sound sits in the word. Click a word to hear it. These are recorded as separate words, not cut out of the sentences above — a word cut from a sentence carries its neighbours' transitions and always sounds muddier. All three groups use the same speaker, so what changes between them is position, not voice.
Word-initial8
Word-medial8
Corpus and acoustic measurements
Click any word in a sentence to play from that word. The pale band over the waveform marks the measured boundary of the target word.
Per-token data
Every number below is measured directly from the audio, not hand-annotated.
| Token | Accent | Sex | Target | Word length | Target consonant duration | F0 | Pitch range | HNR |
|---|
Where these numbers come from
Every claim on this page comes from measuring these 12 recordings. None is written from experience.
Corpus: 12 Spanish sentences generated by a speech engine with two Peninsular voices (one male, one female). Five of the groups are minimal pairs: same vowel, same sentence position, only the voicing of the initial consonant differs. Both members of a pair are read by the same voice, so the difference cannot come from timbre.
Word and phoneme boundaries: MMS_FA forced alignment, with word and phoneme boundaries taken from the same pass. All words across the 12 items came back char-anchored (boundaries anchored to real character boundaries) — none is an estimate.
Voiced ratio: within the middle 80% of the target consonant, one frame every 5 ms, using the peak of the normalised autocorrelation — voiced frames repeat periodically (peak > 0.4), noise frames do not (< 0.2). The threshold is 0.3, midway with margin on both sides. A low-frequency (< 400 Hz) energy ratio is reported alongside as a cross-check.
Why the edges are skipped: the first version measured the whole segment, and the voiceless p in pino came out at 80%. The boundary was correct; the segment edges had picked up voicing from the preceding l. Trimming 10% at each end brought all five pairs to 0% / 100%.