7 Using conversation tools to add meaning and usability
This chapter focuses on making voice applications feel more natural, meaningful, and usable by aligning responses with how people actually speak. It introduces conversation tools that shape both the words and the delivery: discourse markers that guide flow and understanding, and Speech Synthesis Markup Language (SSML) features that control pauses, emphasis, pronunciation, and even embedded audio. The overall aim is to reduce robotic-sounding exchanges, improve clarity, and signal intent and state to users so interactions feel smoother and more human.
A major theme is the power of discourse markers—small words and phrases like “okay,” “so,” “by the way,” “first,” “next,” and “finally”—to connect thoughts, acknowledge understanding, indicate topic shifts, and show progression. Used well, they soften abrupt dialog, add melody, and make prompts friendlier and easier to follow; used poorly, they sound stilted (for example, “otherwise”) or professorial (“indeed,” “to sum up”). The chapter advises choosing markers that match spoken language and the product’s persona, testing them aloud, and being mindful of context and sequencing since markers can subtly change meaning depending on where a response appears. Even simple acknowledgments like “alright” can reassure users they’ve been heard.
SSML is presented as the toolkit for controlling how things are said: breaks and pauses to pace information and create suspense; prosody (rate, pitch, volume) to convey emphasis, excitement, concern, or friendliness; whisper effects for special cases; and pronunciation controls like say-as, role-based w (on Alexa), and phoneme for difficult words, names, numbers, times, and addresses. The chapter notes platform nuances (for example, Google’s support for semitones and lack of the w tag) and recommends focusing SSML polish where analytics show high-frequency prompts. It also covers embedding short audio for branding, structure, or handoffs to recorded voices, with cautions on limits, formatting, and not overusing effects. Combined thoughtfully, these techniques make responses clearer, more expressive, and easier for users to understand and act on.
Summary
- Discourse markers are short words or phrases that provide both conversation flow and meaning.
- SSML is the mechanism through which developers can change the pronunciation or prosody of speech.
- Embedded audio can allow for skill differentiation through sound clips or voices that diverge from the default Alexa voice.
FAQ
What are discourse markers and why do they matter in voice applications?
Discourse markers are small words or phrases (like “okay,” “so,” “by the way”) that connect ideas, signal flow, and add nuance. In VUIs they make responses sound more natural, help users track topic shifts, and convey intent (e.g., sequencing, contrast, clarification) beyond the literal words.Which discourse markers should I prefer for conversational speech, and which should I avoid?
Prefer conversational markers like “okay,” “so,” “by the way,” “first/next/finally,” and “alright.” Avoid overly formal or stilted ones like “therefore,” “in conclusion,” “equally,” or “otherwise,” unless they fit a deliberate persona. Replace “otherwise” with friendlier phrasing (e.g., “If you need something else, hold on…”).How can SSML pauses improve understanding, and what tags should I use?
Use pauses to add suspense, emphasize information, give listeners processing time, or signal “work” being done. Use <p> and <s> for paragraph/sentence pacing and <break> with strength or time (e.g., time="300ms") for precise control. Pauses can also spotlight command phrases and improve teaching moments.What is prosody in SSML and how do rate, pitch, and volume affect the experience?
Prosody controls how speech is delivered: rate (speed), pitch (tone), and volume (loudness). Faster and slightly higher pitch can sound friendlier or excited; slower and lower can convey seriousness or disappointment. Adjust these to add emphasis, emotion, and clarity, especially for frequently spoken lines.Are there platform differences when tuning prosody?
Both Alexa and Google Assistant support rate, pitch, and volume. Alexa accepts percentages and keywords; Google also supports pitch changes via semitones (e.g., +2st). Always test per platform—what sounds emphatic on Alexa may need different values on Google for the same effect.When should I use amazon:effect and is whispering a good idea?
amazon:effect name="whispered" simulates a whisper on Alexa. It can be useful for stylistic emphasis, but it often sounds mismatched or distressed versus normal speech. If you only need quieter delivery, prefer adjusting volume with prosody; reserve whispering for clear, intentional moments.How do I control pronunciation with SSML (w, say-as, phoneme)?
Use <w role="…"> for parts of speech or alternate senses (Alexa only) to disambiguate words like “wind/read.” Use <say-as interpret-as="…"> for numbers, characters, dates, times, units, phone numbers, interjections, or expletive beeps. Use <phoneme> with IPA or X-SAMPA for exact pronunciations (Alexa only) when other methods fail.What are key SSML differences between Alexa and Google Assistant?
- Google doesn’t support <w> or <phoneme> (phoneme content may be dropped).- Google extends dates/times with detail/format controls and pluralizes units automatically.
- Google has a non-standard <par> for parallel audio; Alexa doesn’t speak over embedded audio.
Always verify behavior on each platform.
Voice Applications for Alexa and Google Assistant ebook for free
