All guides

Guide

How do you prompt an AI for brand voice and tone in video ad voiceovers?

, 5 min read

You can prompt an AI for brand voice and tone in video ad voiceovers by using natural language prompts or Speech Synthesis Markup Language (SSML) tags. Google Cloud's AI technologies, such as Cloud Text-to-Speech and Gemini-TTS, offer tools to create unique, natural-sounding speech that represents your brand.

Illustration for How do you prompt an AI for brand voice and tone in video ad voiceovers?

The short version

  • Use natural language prompts with Gemini-TTS for granular control over style, tone, pace, and emotional expression.
  • Employ Speech Synthesis Markup Language (SSML) tags for precise control over pauses, pronunciation, and audio formatting.
  • Select specific AI voice models like WaveNet or Chirp 3: HD, each offering distinct qualities in human-like speech.
  • Adjust pitch up to 20 semitones to personalize the selected voice.
  • Create a unique voice to represent your brand across all customer interactions.

What is AI voice synthesis for brand voice?

AI voice synthesis converts text into natural-sounding speech using advanced AI technologies. This capability allows organizations to create a unique voice that represents their brand across all customer touchpoints, rather than using a generic voice shared with other entities. Google Cloud's Text-to-Speech (Cloud TTS) generates audio with natural, human-like quality, making the speech sound like a real person.

The voices offered by Cloud TTS differ based on the synthetic speech technology used to create the machine model of the voice. This variety enables brands to select or customize a voice that aligns with their desired identity, ensuring consistency and recognition in their video ad voiceovers.

How do you control voice characteristics with natural language prompts?

Google Cloud's Gemini-TTS is an evolution of Cloud TTS technology that provides granular control over generated audio using text-based prompts. With Gemini-TTS, you can synthesize single or multi-speaker speech and precisely dictate style, accent, pace, tone, and emotional expression through natural-language prompts. This allows for adapting the delivery within a conversation, steering it to adopt specific accents, and producing a range of tones and expressions, including a whisper.

Natural language prompts also enable voices to perform with specific emotions and produce accents when requested. Enhanced pace and pronunciation control helps ensure greater accuracy in pronunciation, even for specific words. When using Gemini-TTS, the text field and the prompt field can each be at most 4,000 bytes. The output audio can be around 655 seconds at most, and speaker aliases must consist solely of alphanumeric characters, excluding whitespace.

How do you use SSML for precise speech control?

Speech Synthesis Markup Language (SSML) allows for more customization in your audio response by providing details on pauses and audio formatting. You can send SSML in your Cloud Text-to-Speech request to add pauses, format numbers, dates, and times, and provide other pronunciation instructions. Cloud TTS supports a subset of available SSML tags.

Specific SSML tags offer detailed control. For instance, the `<phoneme>` tag can produce custom pronunciations of words inline. The `<voice>` tag allows you to use more than one voice within a single SSML request. It is important to wrap full sentences with `<s>...</s>` tags, especially if they contain SSML elements that change prosody, such as `<audio>`, `<break>`, `<emphasis>`, `<par>`, `<prosody>`, `<say-as>`, `<seq>`, and `<sub>`. The `<prosody>` tags should only be used around a full sentence.

What are the different AI voice models and their capabilities?

When sending a synthesis request, you must specify a voice to start. Google Cloud offers different voice models, each with distinct characteristics. WaveNet models, for example, are trained using raw audio samples of actual human speech. As a result, these models generate synthetic speech with more human-like emphasis and inflection on syllables, phonemes, and words, making the speech sound very natural.

Another option is Chirp 3: HD voices, which are driven by technology that captures nuances in human intonation, making conversations more engaging. These voices are available 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 the A-Law audio encoding.

How do you adjust pitch and other audio parameters?

You can personalize the pitch of your selected voice using pitch tuning. This allows you to adjust the pitch up to 20 semitones more or less than the default setting, providing flexibility to fine-tune the vocal delivery for your brand's specific tone. This control helps ensure the voiceover matches the desired emotional impact and brand identity.

It is important to note that not all voice models support all audio parameters. For example, Chirp 3: HD voices do not support pitch-audio parameters, meaning pitch adjustments are not available for these specific voice types. Always check the capabilities of the chosen voice model when planning your audio customization.

Common questions

What is Gemini-TTS?
Gemini-TTS is the latest evolution of Google Cloud's Text-to-Speech technology. It moves beyond natural-sounding speech to provide granular control over generated audio. Using natural language prompts, you can precisely dictate style, accent, pace, tone, and emotional expression for single or multi-speaker speech, from short snippets to long-form narratives.
Can I use multiple voices in a single voiceover?
Yes, you can use more than one voice in a single SSML request by employing the `<voice>` tag. This allows for dynamic voiceovers with different speakers or to introduce variety within a single narrative.
What are SSML tags used for?
SSML tags are used to customize your audio response by providing specific instructions for speech synthesis. They allow you to add pauses, format numbers, dates, and times, control pronunciation, and manage emphasis. SSML tags offer a detailed way to fine-tune the delivery of your text-to-speech output.
What is the difference between WaveNet and Chirp 3: HD voices?
WaveNet models are trained on raw audio samples of human speech, resulting in synthetic speech with human-like emphasis and inflection. Chirp 3: HD voices capture nuances in human intonation and come in 30 distinct styles. However, Chirp 3: HD voices do not support SSML input, speaking rate, or pitch-audio parameters, unlike some other models.
What are the length limits for Gemini-TTS prompts and audio output?
With Gemini-TTS, the text field and the prompt field can each be at most 4,000 bytes. The output audio generated from these prompts can be around 655 seconds at most.

Primary sources

This article was researched and written by an AI model and published automatically. It was written from sentences that were checked to appear on the pages below, but a person has not reviewed the text. If something is wrong, tell us at leetcv@darthwares.com and we will correct it.

Turn a script into a video ad

50 free credits. Paste a script, review the storyboard, and render a 30 to 60 second ad.