How to Record a School Robocall

The bandwidth row is the one that changes behaviour. A voice call is narrowband and mono: everything under about 300 Hz and over about 3.4 kHz is thrown away. This is why a recording that sounded rich and authoritative in the conference room sounds thin and mumbled on a parent's phone, and why clear consonants matter far more than a pleasant voice.

Pacing, and the stopwatch method

Read the script aloud, standing, with a stopwatch, twice. Note the time. That is your rate, and it is probably faster than you think under recording conditions and slower than you think when you are nervous.

Word budgets for 60 seconds: 140 words at 140 wpm, 150 at 150, 160 at 160. Use 140 for anything emergency-related and 150 for routine operations. Anything above 160 wpm is difficult to follow on a narrowband line and materially harder for a listener whose first language is not English.

Four pacing habits that survive the phone line:

• Pause a full beat between the headline and the detail. On a phone this reads as structure. Without it the message is a wall.
• Slow down for numbers. Times, dates, addresses and phone numbers should be read at roughly half your normal rate, with a pause before and after.
• Say phone numbers in groups with pauses, and say them twice if the message asks people to call.
• Do not smile through an emergency message. The convention that a smile improves a recording comes from customer service and is wrong for a lockdown.

Human voice or text-to-speech

The honest answer is that both are right, for different messages.

Use a human voice for: any emergency or incident; any message where the district's credibility is the point; the all-clear; anything about a death; the first message of the school year; and any message a family may hear twice and remember.

Use text-to-speech for: overnight and pre-dawn sends where nobody is available to record; high-frequency operational messages such as daily absence notifications; languages where you have no fluent reader on staff, because a competent synthetic read in Vietnamese is better than a phonetic attempt by an English speaker; and messages assembled from data where a human read would be error-prone.

Modern synthetic voices are good enough that most listeners will not identify them on a phone line, which is exactly why the distinction should be a policy rather than a habit. Districts drift toward text-to-speech for everything because it is faster, and then find that the lockdown message sounded like the book fair message.

One rule worth writing into the policy: the superintendent records the emergency messages. Not the communications director, not a synthetic voice. When something serious has happened, the voice families hear should be the one accountable for it, and the recording should be made freshly rather than pulled from a library — a pre-recorded emergency message will eventually be sent for the wrong incident.

Names, and when to leave them out

Names are where recordings do the most avoidable damage. Three cases:

• Staff names. Ask the person how they say their own name, record the answer, keep a pronunciation list in the communications folder. Getting a colleague's name wrong on a message to twelve thousand households is a small humiliation delivered at scale.
• School and place names. Local pronunciations of street and school names are frequently not what spelling suggests, and a superintendent new to the district will get one wrong. Have someone local listen before it goes out.
• Family and student names. A broadcast should not contain one. If a message concerns an individual child, it is a call to that family, not a broadcast — and the merge-field version of this is worse, because a recorded message that splices in a synthesised student name will mangle a proportion of them and each mangling lands with the family least likely to forgive it.

Where a name would be mispronounced and there is no time to check, use the role: "the school office", "your child's teacher", "the attendance office". A correct role beats a mangled name every time.

Bilingual recordings

The single most common defect in district voice broadcasts: sixty seconds of English followed by sixty seconds of Spanish, and the Spanish-speaking listener hangs up at fifteen seconds because they have no reason to believe anything is coming for them.

Three workable patterns:

• Separate sends by language of record. Best when your language data is good. Each family hears one message in their language, the call is half as long, and the cost per family is the same. This requires reliable language-of-record data, which many districts do not have.
• Bridged bilingual, one call. Open with a five-second bilingual bridge before the English begins: "This message is in English and Spanish. Este mensaje está en inglés y español. El español comienza en un minuto." That one sentence is the difference between a listener waiting and a listener hanging up. Then English, then a two-second gap, then Spanish.
• Spanish first. In a district where Spanish-preference households are the majority, put Spanish first and bridge in English. Districts rarely consider this and it is often correct.

Use two voices, one per language, each a fluent speaker. A single reader working through a language they do not speak is worse than a synthetic voice, and families can tell.

Questions people actually ask

Related