Acoustic analysis of the neutral tone in Mandarin

Philippe Martin, Jun LiInterspeech 2007

TL;DR

This paper investigates the acoustic properties of the Mandarin neutral tone (TO), which arises in specific tonal sandhi contexts, particularly with sequences of third tones. By analyzing fundamental frequency (Fo) glissando and duration, the authors propose that the neutral tone is acoustically distinct from the third (T3) and fourth (T4) tones, primarily through its less sharp Fo fall and absence of a rising contour.

What Problem Does This Paper Address?

Mandarin Chinese is a tonal language, meaning the pitch contour of a syllable carries lexical meaning. While isolated tones have well-defined acoustic shapes (e.g., high, rising, falling), these contours can change when tones appear in sequence, a phenomenon known as tonal sandhi. A particularly complex case involves sequences of third tones (T3), which can result in a "neutral tone" (TO) instead of the expected T2. Previous research has offered differing explanations for the acoustic realization of TO, with some suggesting it's an interpolation of surrounding tones. This paper aims to clarify the specific acoustic features that distinguish the neutral tone from other tones, particularly T3 and T4, in its natural sandhi context.

Methods

The researchers recorded 60 sentences containing sequences of three or more T3 tones, spoken by native Mandarin speakers. They then used acoustic analysis software (WinPitchPro) to extract fundamental frequency (Fo) contours, intensity, and spectrograms. A panel of listeners identified which syllables were perceived as neutral tones. The core of the analysis focused on two acoustic parameters:

  • Glissando: This refers to the rate of change in fundamental frequency over time. The paper specifically looks at the glissando value of the Fo contour, comparing it to a "glissando threshold" – the minimum Fo change needed for a listener to perceive a pitch movement rather than a static tone.
  • Duration: The length of the syllable was also measured.

The authors also conducted a synthesis experiment where they modified the Fo contours of perceived neutral tones to see if they could be made to sound like T3 or T4, and vice-versa.

Datasets

The study utilized a corpus of 60 newly recorded sentences designed to elicit neutral tones in Mandarin. These sentences were constructed to include various combinations of tones and syllable structures, particularly focusing on sequences of third tones. The sentences were spoken by 6 native Mandarin speakers. The paper also references existing literature and perceptual studies that used synthesized data or other corpora.

Key Insights and Ideas

  • Neutral tone is acoustically distinct: The neutral tone isn't simply an average of surrounding tones. It possesses unique acoustic characteristics that differentiate it from T3 and T4.
  • Glissando is key, but not the sole determinant: While the glissando (Fo fall rate) of the neutral tone is generally lower than that of T4, and it lacks the rising contour of T3, the standard deviation of glissando values was large, suggesting other factors are at play.
  • Absence of a rising contour is crucial: The neutral tone contrasts with T3 by not having a final rising bend in its Fo contour, even if that bend is very slight.
  • Sharpness of Fo fall differentiates from T4: The neutral tone's Fo fall is less sharp than that of T4. The paper suggests a threshold around 20 ST/s for neutral tones versus 30 ST/s for T4.
  • Vowel quality might play a role: The authors acknowledge that vowel quality, not extensively analyzed in this study, could influence the perception and production of neutral tones.
  • Creaky phonation can occur: In some instances, particularly at the end of sentences, neutral tones (and even T3) can exhibit creaky phonation, which might be an additional acoustic cue.

Key Results

  • The study found that neutral tones often exhibit a complex Fo pattern, with an initial fall followed by a flat or slightly falling bend.
  • Perceptual tests on the recorded corpus indicated that approximately 30% of T3 sequences were perceived as neutral tones (T3 -> TO), and a similar percentage were perceived as T2 (T3 -> T2).
  • The synthesis experiment showed that modifying a neutral tone's Fo contour to be sharper could make it sound like T4, and adding a rising bend could make it sound like T3. Conversely, a sharp falling contour (like T4) could be perceived as neutral if the fall was not too steep.

Limitations and Future Work

  • Listener variability: The authors note that the stability of listener judgments in identifying neutral tones can be a source of variability in the results.
  • Influence of vowel quality: The study did not deeply investigate the impact of different vowel qualities on neutral tone realization, suggesting this as a direction for future research.
  • Creaky phonation: While observed, the precise role and frequency of creaky phonation in neutral tone perception requires further study.
  • Broader context: The study focused on specific T3 sandhi contexts. Future work could explore neutral tone realization in other tonal environments or across different dialects of Mandarin.

Summary updated

BibTeX
@inproceedings{martin2007acoustic,
  author = {Philippe Martin and Jun Li},
  title = {{Acoustic analysis of the neutral tone in Mandarin}},
  booktitle = {Interspeech},
  year = {2007},
}
Figure