This week Amazon Alexa certified the skill we’ve been working on with Women in Voice from around the world. Freedom speaks features stories, poems and music honoring Iranian women and brings attention to their inspiring resilience.
Just say “Alexa, open Freedom Speaks!” to make history come alive.
Through voice AI, we bring the words of iconic Iranian women from the past to life. A storyteller for the modern world, Freedom Speaks can inspire others to leverage it as a tool for freedom of expression and provide a platform for voices from different cultural and ethnic backgrounds. Sign up here, to join our launch on February 9 at 2pm ET.
#FreedomSpeaks Voice AI: Launch Event - In Honor of Iranian Women
(in partnership with Davar Ardalan, Women in AI USA and Women in Voice) Don't miss the chance to celebrate freedom and…
Iranian women have been defying odds and inspiring us for centuries. Join us in celebrating their courage, diversity, and strength.
Freedom speaks was produced (in alphabetical order) by Carrie Jaquith, Ezinwa Amadi, Iran Davar Ardalan, Kimberly Mishra, Maddy Apple, Nikki McLay, Paris Golab, Sahar Mazloom, Tali Weinberg, and Tina Soleimanifar. Our February 9th launch event is in partnership with Women in Voice and Women in AI together with Bhuva Subram.
This evening, I spent some time with ChatGPT. I was curious what I’d learn about the lack of Persian language support in many voice and text-to-speech applications. It’s an important language to include for global diversity — especially now when so many voices in Iran are silenced.
ChatGPT breaks it down pretty well and our conversation is super informative.
- Persian is not as widely spoken as some other languages, so there may be less demand for it.
- Developing support for a new language can be a complex and time-consuming process, and companies may prioritize adding support for languages that are more widely spoken.
- There may also be a lack of high-quality data and resources available for training AI models on the Persian language.
We also talk about why we need to fix this. It will improve access to information and preserve the Persian language.
“AI can be used to improve access to information in the Persian language by creating tools such as machine translation, text-to-speech, and language processing tools.
With the help of AI, it will be easier to preserve the Persian language by creating accurate and up-to-date resources such as dictionaries, corpora, and language models.” — ChatGPT
Ardalan: What kind of data should we feed you?
ChatGPT: To make me smarter in the context of the Persian language and culture, there are a few types of data that could be useful to feed me:
- High-quality text data: This includes a wide range of texts such as books, articles, websites, and social media posts. These texts should be diverse and representative of the different styles and registers of the Persian language, as well as the different topics and genres found in Persian literature and culture.
- Audio and video data: This includes speech and music recordings, as well as videos. This data can help me to learn the nuances of the spoken Persian language, as well as its rhythm, intonation, and accent.
- Structured data: This includes data that is organized in a structured format, such as databases, spreadsheets, and annotated corpora. This type of data can be used to train models for specific tasks such as named entity recognition, sentiment analysis, and machine translation.
- Image and graphic data: This includes images and graphic materials such as Persian paintings, calligraphy, and architecture. These materials can help me to understand the visual and aesthetic aspects of Persian culture and art.
Ardalan: Did you know that I just helped create an Alexa Skill called Freedom Speaks? It includes stories and poems about remarkable Iranian women. We did this because we want to show the lack of Persian datasets and how we need to fix this.
ChatGPT: “I am not aware of any specific Alexa Skill called Freedom Speaks but it is great to hear that you are working on creating a dataset of Persian stories and poems about remarkable Iranian women. It is important to create diverse and representative datasets to help train models like mine to better understand and generate text in the Persian language.
Such an Alexa skill will be a great addition to the Persian language dataset, and it will help raise awareness about the importance of having a diverse dataset. Creating a diverse dataset is key to making sure that the model understands the complexity and nuances of the Persian language and culture. It’s also important to ensure that the data is well-curated, high-quality, and free from biases. This can help to improve the performance of models like mine and make them more useful for a wider range of tasks.”
“It’s great to see that you are working to help improve the state of the art in natural language processing for the Persian language, keep up the good work!” — ChatGPT
With Freedom Speaks we want to also highlight the absence of Persian speech in training datasets. We want to explore how Voice AI can be improved for the Persian language in the service of empowering Iranian women.
Go ahead try it and tell us what you think. Just say “Alexa, open Freedom Speaks!”