Example code to generate WAV audio from text using Gemini TTS and present it in a Shiny audio preview modal

[diegoperoni]

This issue provides working example code to generate WAV audio from text using Gemini TTS and present it in a Shiny audio preview modal.

Example Code

options(ellmer_timeout_s = 900) # for long audio

gemini_tts = function(text, voice="charon", model="gemini-2.5-flash-preview-tts", api_key=NULL) {
  api_key = api_key %||% Sys.getenv("GOOGLE_API_KEY")
  if (is.null(api_key) || api_key == "") {
    stop("Google API key not found. Set GOOGLE_API_KEY or pass api_key as a parameter.")
  }
  # Internal function to create a WAV header from raw PCM
  create_wav_header = function(data_length, sample_rate = 24000, bits_per_sample = 16, channels = 1) {
    byte_rate <- sample_rate * channels * bits_per_sample / 8
    block_align <- channels * bits_per_sample / 8
    header <- raw(44)
    # "RIFF" chunk
    header[1:4] <- charToRaw("RIFF")
    size <- data_length + 36
    header[5:8] <- writeBin(as.integer(size), raw(), size = 4, endian = "little")
    header[9:12] <- charToRaw("WAVE")
    # "fmt " subchunk
    header[13:16] <- charToRaw("fmt ")
    header[17:20] <- writeBin(16L, raw(), size = 4, endian = "little")
    header[21:22] <- writeBin(1L, raw(), size = 2, endian = "little")
    header[23:24] <- writeBin(as.integer(channels), raw(), size = 2, endian = "little")
    header[25:28] <- writeBin(as.integer(sample_rate), raw(), size = 4, endian = "little")
    header[29:32] <- writeBin(as.integer(byte_rate), raw(), size = 4, endian = "little")
    header[33:34] <- writeBin(as.integer(block_align), raw(), size = 2, endian = "little")
    header[35:36] <- writeBin(as.integer(bits_per_sample), raw(), size = 2, endian = "little")
    # "data" subchunk
    header[37:40] <- charToRaw("data")
    header[41:44] <- writeBin(as.integer(data_length), raw(), size = 4, endian = "little")
    header
  }
  # Build URL
  base_url = "https://generativelanguage.googleapis.com/v1beta/models"
  url = paste0(base_url, "/", model, ":generateContent")
  # Create and execute request
  req = httr2::request(url) |>
    httr2::req_url_query(key = api_key) |>
    httr2::req_timeout(60*15) |>
    httr2::req_body_json(list(
      contents = list(list(
        parts = list(list(text = text))
      )),
      generationConfig = list(
        responseModalities = list("AUDIO"),
        speechConfig = list(
          voiceConfig = list(
            prebuiltVoiceConfig = list(
              voiceName = voice
            )
          )
        )
      )
    ))
  resp = httr2::req_perform(req)
  json_resp = httr2::resp_body_json(resp)
  # Extract base64 audio
  audio_base64 = json_resp$candidates[[1]]$content$parts[[1]]$inlineData$data
  if (is.null(audio_base64)) {
    stop("No audio received from the API response")
  }
  # Decode and create WAV
  audio_pcm  = base64enc::base64decode(audio_base64)
  wav_header = create_wav_header(length(audio_pcm))
  # Return complete raw WAV
  c(wav_header, audio_pcm)
}

audio_preview_modal = function(wav_b64, title = "Audio Preview (TTS)") {
  shiny::modalDialog(
    title = title,
    size = "m",
    easyClose = TRUE,
    bslib::card(
      bslib::card_header(class = "bg-dark text-light", title),
      bslib::card_body(
        style = "background-color: #000000; color: #e0e0e0;",
        shiny::tags$audio(
          controls = NA,
          autoplay = NA,
          style = "width: 100%;",
          shiny::tags$source(
            src = paste0("data:audio/wav;base64,", wav_b64),
            type = "audio/wav"
          )
        ),
        shiny::tags$hr(),
        shiny::tags$p(class = "text-muted", "If the audio doesn't start, press Play.")
      )
    ),
    footer = shiny::modalButton("Close")
  )
}

# MAIN
wav_data = gemini_tts(text = "Hi, I'm Google Gemini Text to Speech.")
wav_b64 = base64enc::base64encode(wav_data)
shiny::showModal(audio_preview_modal(wav_b64))

---

Instructions:

• Make sure you have your Google API key properly set (see code comments).
• Adjust voice and model parameters as needed for your application.
• Use the audio preview modal in your Shiny app to allow users to listen and download generated audio.