Speech-to-Text Options Guide

This guide explains the available Speech-to-Text options from a user and outcome perspective.

These settings control how speech is converted into text, how readable the transcript is, and whether additional analysis such as topics, sentiment, entities, or intents is produced.

Where a setting can be changed by the client, it means the option can be altered for a specific connection or request. Where it cannot be changed, the system uses the configured service default.

An Overview of the Options

language

This option can be changed or overridden by settings

The default language is en-GB.

What this does

Sets the expected spoken language for the audio. This helps the speech recognition system choose the most appropriate language model and vocabulary.

When to use this

Use this when you know the main language being spoken.

Use UK English for UK calls, US English for US callers, or another supported language when the call is expected to be in that language.

interim results

This option can be changed or overridden by settings

The default is is false.

What this does

Controls whether partial transcript results are returned while someone is still speaking.

When enabled, the application can show live text before the final transcript is confirmed.

When to use this

Use interim results when you want a fast delivery of the words, but note that the final transcript can be different in more cases than you might expect! You may need to replace partials with final transcripts on the fly depending on your application.

Example outcomes

• Display words on screen as the caller speaks.
• Power real-time call assistance.
• Show live captions before the final transcript is complete.

endpointing

This option can be changed or overridden by settings

What this does

Controls how quickly the speech recognition system decides that a speaker has paused or finished speaking.

A lower value can make the system respond faster. A higher value can reduce the chance of cutting off a speaker too early.

When to use this

Use this when you need to tune the balance between responsiveness and accuracy.

Example outcomes

• Use a shorter endpoint for fast, interactive voice applications.
• Use a longer endpoint where speakers pause naturally between phrases.
• Disable or reduce endpointing sensitivity where premature end-of-speech detection causes problems.

detect language

This option can be changed or overridden by settings

The default value is false.

What this does

Asks the system to identify the dominant spoken language in the audio.

This is different from setting the language directly. Instead of telling the system what language to expect, this asks the system to detect it.

When to use this

Use this when the caller’s language may not be known in advance.

Example outcomes

• Identify whether a caller is speaking English, Spanish, French, or another supported language.
• Route or tag calls based on detected language.
• Support multilingual services where the expected language is uncertain.

punctuate

This option cannot be changed or overridden by settings

What this does

Adds punctuation and capitalization to the transcript.

Without punctuation, transcripts can be harder to read because they may appear as a continuous stream of words.

smart format

This option can be changed or overridden by settings

The default valie is false.

What this does

Applies intelligent formatting to recognized text.

This can improve how dates, times, currencies, phone numbers, and similar values appear in the transcript.

When to use this

Use this when the transcript should be clean, readable, and useful for business records.

Example outcomes

• Convert spoken dates into a cleaner date format.
• Improve formatting of times, prices, and numbers.
• Make transcripts more suitable for CRM, reporting, or review.

numerals

This option can be changed or overridden by settings

The default value is false.

What this does

Converts spoken numbers into numeric form.

For example, spoken “one hundred and twenty five” may be shown as 125.

Example outcomes

• Make account numbers easier to read.
• Improve capture of quantities, prices, dates, or reference numbers.
• Reduce manual cleanup of transcripts.

measurements

This option can be changed or overridden by settings

The default value is false.

What this does

Converts spoken measurements into abbreviated or formatted measurement values where supported.

When to use this

Use this when calls are likely to include dimensions, distances, weights, or other measurements.

Example outcomes

• Convert spoken measurement phrases into clearer written forms.
• Improve technical, logistics, medical, engineering, or field-service transcripts.
• Make measurement-heavy conversations easier to scan.

filler words

This option cannot be changed or overridden by settings

What this does

Controls whether the following filler words are included in the transcript.

• uh
• um
• mhmm
• mm-mm
• uh-uh
• uh-huh
• nuh-uh

When to use this

Use this when you care about whether natural speech hesitations should appear in the transcript.

profanity filter

This option can be changed or overridden by settings

The default value is false.

What this does

Filters recognized profanity from the transcript.

When to use this

Use this when transcripts may be displayed to users, stored in business systems, or shared more widely.

This might only include major swear words and not slang words that some may find offensive.

Example outcomes

• Reduce offensive language in visible transcripts.
• Make transcripts more suitable for customer records.
• Support moderation or compliance requirements.

detect entities

This option can be changed or overridden by settings

The default value is false.

See refernce of supported types at the bottom of this document.

What this does

Identifies structured information in the audio or transcript.

Entities may include names, phone numbers, dates, addresses, and similar values.

When to use this

Use this when the transcript should produce useful data, not just text.

Example outcomes

• Extract phone numbers mentioned during calls.
• Identify dates, names, addresses, or reference information.
• Support CRM capture, workflow automation, or reporting.

diarize

This option cannot be changed or overridden by settings

The default value is false.

We are currently using a single stream for each channel, meaning this feature is not required.

What this does

Attempts to identify speaker changes in a single audio stream.

The transcript may assign words or sections to different speaker numbers.

When to use this

Use this when more than one person is speaking and the audio is not separated into different channels.

mode

This option can be changed or overridden by settings

What this does

Sets the broad operating domain for the recognition or analysis.

Supported modes are typically:

general
medical
finance

The default mode is usually general.

When to use this

Use this when calls belong to a specific specialist domain.

Example outcomes

• Use general for normal business calls.
• Use finance where financial terms are common.
• Use medical where medical vocabulary is expected.

replace

This option can be changed or overridden by settings

The default value is for no replacements.

What this does

Searches for terms or phrases and replaces them in the transcript output.

This is useful for known corrections, especially product names, brand names, acronyms, or terms that are often misheard.

When to use this

Use this when you know a predictable transcription error should always be corrected.

Multiple replacement options can be supplied.

Example outcomes

• Replace “dog” with cat.
• Correct commonly misrecognized product names.
• Standardize terminology in final transcripts.

redact

This option can be changed or overridden by settings

The default value is for no redactions.

What this does

Redacts sensitive information from transcripts where supported.

When to use this

Use this when transcripts may contain information that should not be stored or displayed in full.

Multiple redaction options can be supplied. See table at the bottom of thsi document for supported entities.

Example outcomes

• Remove personally identifiable information.
• Reduce exposure of sensitive data in stored transcripts.
• Support privacy, security, or compliance requirements.

key terms

This option can be changed or overridden by settings

The default value is for no key terms.

What this does

Provides key term prompting to help the model recognize important words, names, brands, or specialist phrases.

For Nova-3, this is usually the preferred way to help the system recognize important terminology.

When to use this

Use this when specific terms are especially important to the transcript.

Multiple key terms can be supplied:

Example outcomes

• Improve recognition of product names.
• Help the model understand business-specific vocabulary.
• Support more accurate transcription of names, acronyms, and specialist terms.

Supported Entity Types

This section lists the supported entity types grouped by category.

PII: Personally Identifiable Information

PHI: Protected Health Information

PCI: Payment Card Industry

Other Entities