TL;DR
This paper investigates how speakers use subtle changes in their vocal fold vibration (laryngeal voice quality) to signal emphasis (focus) in speech, going beyond traditional pitch and loudness cues. They found that contrary to expectations, focused speech might be produced with a less tense, or breathier, voice quality, and this effect is more pronounced in certain word positions.
What Problem Does This Paper Address?
When we speak, we often emphasize certain words or phrases to convey specific meanings, a phenomenon known as "focus." While we intuitively understand this, precisely how speakers achieve this emphasis acoustically is complex. Researchers know that changes in fundamental frequency (pitch) and intensity (loudness) are key cues. However, this paper delves into a less-explored aspect: laryngeal voice quality. This refers to the way the vocal folds vibrate, which can range from a "pressed" (tense) to a "breathy" (lax) sound. The authors aim to understand if and how these laryngeal voice quality changes contribute to signaling linguistic focus, particularly in Finnish, and how these changes relate to acoustic cues like pitch.
Methods
The researchers employed inverse filtering (IF), a technique that estimates the glottal airflow (the air passing through the vocal folds) by removing the influence of the vocal tract. This allows them to analyze the raw vibration of the vocal folds. They used an Iterative Adaptive Inverse Filtering (IAIF) algorithm to estimate this glottal flow.
To quantify voice quality from the estimated glottal flow, they used the Normalized Amplitude Quotient (NAQ). NAQ is a parameter that captures the shape of the glottal closing phase. A low NAQ value indicates a pressed voice quality (short closing phase), while a high NAQ value suggests a breathy voice quality (long closing phase). They also measured fundamental frequency ($f_0$).
The analysis was performed using the TKK Aparat software package, which is designed for voice inverse filtering and parameterization. They used mixed-effects models to analyze the data, accounting for individual speaker variations and other factors.
Datasets
The study used a custom dataset of simple declarative Finnish sentences. These sentences were designed to elicit three types of focus:
- Narrow focus: Emphasis on a specific word (subject or object).
- Broad focus: General emphasis on the entire utterance.
- No focus: Neutral or unfocused speech.
The sentences featured two names as subjects and objects, and the target words contained either short or long vowels ([a, æ, e, o]) surrounded by voiceless consonants to minimize coarticulation effects. A total of 144 sentences were recorded per participant.
Key Insights and Ideas
- Focus is more than just pitch: While pitch ($f_0$) is a primary indicator of focus, laryngeal voice quality (as measured by NAQ) also plays a role, though perhaps in a less intuitive way than initially hypothesized.
- Breathy voice for focus? The study found that focused vowels tended to have higher NAQ values (suggesting a breathier voice quality) compared to broadly focused or unfocused vowels. This contradicts the initial hypothesis that focus would be associated with a more pressed voice quality.
- Word position matters: The effects of focus on both pitch and voice quality were more pronounced in the second word position of the sentence compared to the first.
- Unfocused words are attenuated: Unfocused words, particularly in the second position, showed significantly lower NAQ values, indicating a more pressed voice quality.
- Pitch-driven voice quality: The authors propose that the observed voice quality changes might be a consequence of the vocal fold adjustments needed to produce higher pitches for focus, rather than a direct manipulation of vocal fold tension for voice quality itself. This aligns with theories suggesting higher pitch is associated with smoother glottal pulses.
- Speech economy: The authors speculate that speakers might avoid simultaneously increasing pitch and vocal fold tension (pressedness) for focus to maintain speech economy, leading to a breathier quality at higher pitches.
Key Results
- Fundamental Frequency ($f_0$):
- Higher $f_0$ was observed in the first word position compared to the second.
- Narrow focus resulted in significantly higher $f_0$ than broad focus.
- The "no-focus" condition showed significantly lower $f_0$ in the second position compared to the first.
- Normalized Amplitude Quotient (NAQ):
- NAQ values were higher in narrowly focused words compared to broadly focused or unfocused words.
- NAQ values were significantly lower in the unfocused word in the second position.
- The interaction between focus and word position was significant, with the difference in NAQ being less pronounced in the first position.
- Gender had an effect on NAQ, but vowel type did not.
Limitations and Future Work
- Limited vowel set: The study focused on a specific set of vowels ([a, æ, e, o]), and the results might not generalize to all vowels.
- Read speech: The data was collected from read speech, which may not fully represent spontaneous speech patterns.
- Preliminary findings: The authors acknowledge that their previous study on this topic yielded preliminary results, and while this study extends it, more comprehensive data is needed.
The authors suggest that further research should focus on gathering more comprehensive data on voice source behavior in natural continuous speech and applying robust inverse filtering and parameterization methods. They highlight the TKK Aparat software as a valuable tool for such future investigations.
Summary updated