How do you prompt an AI for different voice styles or accents in video ads?

How do you use natural language prompts for voice styles?

Natural language prompts allow you to precisely dictate the style, accent, pace, tone, and emotional expression of an AI-generated voice. This method uses everyday language to guide the speech synthesis process, making it accessible for marketers and creators. Google Cloud Text-to-Speech, for example, supports these prompts in over 75 locales, enabling broad customization.

With models like Gemini-TTS, you can learn how to control speech synthesis by using natural language prompts to specify the desired style, tone, pace, and emotional expression. This approach simplifies the process of achieving a specific vocal delivery, allowing you to focus on the creative aspects of your video ad script without needing complex technical commands.

What is Speech Synthesis Markup Language (SSML) and how does it work?

Speech Synthesis Markup Language, or SSML, provides a way to add more customization to your audio response when converting text to speech. Instead of simple plaintext, you can embed SSML tags within your script to provide detailed instructions for the AI model. These tags allow you to control elements such as pauses, number and time formatting, delivery, pronunciation, and emotional expression.

For example, Google Cloud Text-to-Speech documentation explains that SSML tags can be used to add pauses, format acronyms, dates, and times, or even censor specific text. The `<voice>` tag allows you to use multiple voices within a single SSML request, provided you specify a compatible voice name or attributes. The `<prosody>` tag customizes the pitch, speaking rate, and volume of text, though it should only be used around a full sentence. You can also use the `<phoneme>` tag for custom pronunciations of words inline, and the `<emphasis>` tag to add or remove emphasis from text.

When using SSML, it is important to avoid using SSML reserve characters in the text that is to be converted to audio. This ensures that the markup is interpreted correctly and does not interfere with the content of your script.

What voice options are available for AI-generated speech?

Google Cloud Text-to-Speech offers a wide selection of voices designed to generate natural, human-like audio. You can choose from over 380 voices available across more than 75 languages and their variants, including Mandarin, Hindi, Spanish, Arabic, and Russian. To begin, you specify a voice when sending a synthesis request to the API.

Different voice types are available for various needs. Chirp 3: HD voices, for instance, are driven by technology that captures nuances in human intonation, making conversations more engaging. These voices come in 30 distinct styles across many languages and are suitable for both real-time and standard applications. However, Chirp 3: HD voices do not support SSML input, speaking rate and pitch-audio parameters, or A-Law audio encoding.

Other options include Studio voices, which are specifically designed for news reading and broadcast content, offering a professional tone. Neural2 voices are based on the same technology used for Custom Voice models. These models generate synthetic speech with more human-like emphasis and inflection on syllables, phonemes, and words, providing a highly realistic output.

How can you create a custom voice for your brand?

To establish a strong brand identity, you can create a unique voice that represents your brand across all customer touchpoints. Google Cloud's Custom Voice feature within its Text-to-Speech API allows you to train custom voice models using your own audio recordings. This means you can develop a voice that is distinct to your organization, rather than using a common voice shared with others.

The process for creating a custom voice is streamlined: you can submit your audio recordings and gain access to the new voice directly in the Text-to-Speech API. It is possible to create personalized voice models with as little as 10 seconds of audio input. Custom Voice Text-to-Speech also provides guidance on audio requirements to help ensure you generate a high-quality custom voice model.

Once a new custom voice model is trained, you can start using it by referencing its model ID in your calls to the Cloud Text-to-Speech API. There is a review process for Custom Voice Text-to-Speech to ensure that each use case aligns with Google's AI Principles and that adequate voice actor consent is obtained. To verify voice actor consent, you may need to submit an audio file of the actor producing a specific sentence chosen by Google Cloud, such as: “I agree that my voice will be used to create a synthetic custom Text-to-Speech voice.” Custom Voice Text-to-Speech is generally available in several languages, including English (US, AU, UK), Spanish (US, Spain), French (France, Canada), Italian (Italy), German (Germany), Portuguese (Brazil), and Japanese (Japan).

Common questions

How many voices and languages does Google Cloud Text-to-Speech offer?

Google Cloud Text-to-Speech offers a selection of over 380 voices across more than 75 languages and variants, including Mandarin, Hindi, Spanish, Arabic, and Russian.

Can I control emotional expression with AI voice generation?

Yes, you can precisely dictate emotional expression, along with style, accent, pace, and tone, using simple natural-language prompts. Models like Gemini-TTS allow for this detailed control over speech synthesis.

What is the purpose of SSML tags in speech synthesis?

Speech Synthesis Markup Language (SSML) tags are used to add customization to your audio response. They provide details on pauses, audio formatting for acronyms, dates, times, abbreviations, or text that should be censored, and allow control over pitch, speaking rate, volume, and emphasis.

How much audio is needed to create a custom voice?

You can create personalized voice models with as little as 10 seconds of audio input using Google Cloud's Custom Voice feature. This allows businesses to train unique voice models with their own recordings.

What are Chirp 3: HD voices?

Chirp 3: HD voices are a type of AI-generated voice driven by technology that captures nuances in human intonation, making conversations more engaging. They are available in 30 distinct styles across many languages and are suitable for real-time and standard applications, though they do not support SSML input or certain audio parameters.