Best Free Speech-to-Text APIs and Open Source Engines: Comprehensive Comparison

Choosing the best speech-to-text API, AI model, or open-source engine to build can be difficult. You need to consider factors such as accuracy, model design, features, support options, documentation, and security. According to AssemblyAI, this post looks at the best free speech-to-text APIs and AI models on the market today, including those that offer free tiers.

Free speech-to-text API and AI models

APIs and AI models are generally more accurate and easier to integrate than open source options. However, APIs and AI models can be expensive to use at scale. For small projects or pilots, many Speech-to-Text APIs and AI models offer free tiers that allow users to leverage the service up to a certain volume. Three popular Speech-to-Text APIs and AI models with free tiers are AssemblyAI, Google, and AWS Transcribe.

Assembly AI

AssemblyAI provides AI models that accurately transcribe and understand speech, allowing users to extract insights from their speech data. It offers cutting-edge AI models such as Speaker Diarization, Topic Detection, Entity Detection, Automated Punctuation and Casing, Content Moderation, Sentiment Analysis, and Text Summarization. AssemblyAI supports virtually all audio and video file formats for easier transcription, and offers two options for speech-to-text conversion: “Best” and “Nano.” The company also offers $50 in credits to get users started.

price

• Free to test on AI Playground, $50 credit when you sign up for API
• Best Voice-to-Text – $0.37/hour
• Voice-to-text Nano – $0.12/hour
• Streaming Voice-to-Text – $0.47 per hour
• Speech Understanding – Varied
• Volume pricing available

merit

• High accuracy
• Various AI models
• Continuous model improvement
• Developer-friendly documentation and SDKs
• Pay-as-you-go and custom plans
• Strict security and privacy practices

disadvantage

• The model is not open source

Google

Google Speech-to-Text offers 60 minutes of free transcription and $300 in free credit for Google Cloud hosting. However, Google will only transcribe files that are already in your Google Cloud Bucket, and requires you to set up a Google Cloud Platform (GCP) account and project.

price

• 60 minutes of free transcripts
• $300 in free credit for Google Cloud hosting

merit

• Free Tier
• Good accuracy
• Supports over 125 languages

disadvantage

• Only conversion of files in Google Cloud Bucket is supported.
• Initial setup can be complicated
• Lower accuracy compared to other APIs

AWS Warrior

AWS Transcribe offers one hour per month free for the first 12 months. Like Google, you need an AWS account and the files must be in an Amazon S3 bucket. AWS Transcribe also offers medical transcription capabilities via the Transcribe Medical API.

price

• 1 hour free per month for the first 12 months
• Depending on usage, the price range varies from $0.02400 to $0.00780.

merit

• Integration into the AWS ecosystem
• Medical Language Transcription
• Good accuracy

disadvantage

• Initial setup can be complicated
• Only transcription of files in Amazon S3 buckets is supported.
• Lower accuracy compared to other APIs

Open source speech transcription engine

Open source Speech-to-Text libraries are completely free and have no usage restrictions. These libraries can provide better data security as they do not require data to be transmitted to third parties. However, they often require significant time and effort to achieve the desired results, especially at scale. Here are some notable open source options:

Deep speech

DeepSpeech is an open source, embedded speech-to-text engine designed to run in real time on a variety of devices. It offers excellent baseline accuracy and is easy to fine-tune and train on custom data.

merit

• Easy to customize
• You can train custom models
• Runs on a variety of devices

disadvantage

• Lack of support
• There is no model improvement outside of custom learning.
• Complex integration into production applications

Remains

Kaldi is a popular speech recognition toolkit in the research community. It provides excellent baseline accuracy and supports custom model training. Kaldi is widely used in production by many companies.

merit

• Good accuracy
• Custom model support
• Active user base

disadvantage

• It’s complicated and expensive to use.
• Use the command line interface
• Complex integration into production applications

Flashlight ASR (formerly Wav2Letter)

Flashlight ASR is an automatic speech recognition (ASR) toolkit from Facebook AI Research. It is written in C++ and uses the ArrayFire tensor library. Flashlight ASR is customizable and provides reasonable accuracy for an open source option.

merit

• Customizable
• Easier to modify than other open source options.
• High processing speed

disadvantage

• It’s very complicated to use
• Pre-trained libraries are not available.
• Continuous data set sourcing is required for training.

Speech Brain

SpeechBrain is a PyTorch-based transcription toolkit that is tightly integrated with Hugging Face, making it easy to access. The platform is well-defined and constantly updated, making it a simple tool for training and fine-tuning.

merit

• Integration with Pytorch and Hugging Face
• Pre-trained models available
• Support for various tasks

disadvantage

• Pre-trained models require customization.
• Lack of extensive documentation

Cook

Coqui is a deep learning toolkit for speech-to-text conversion. It supports multiple languages ​​and provides essential inference and production capabilities. The platform also allows for custom training models to be published and provides bindings for various programming languages.

merit

• Generates a confidence score for a manuscript.
• Large support community
• Pre-trained models available

disadvantage

• No longer updated in Coqui.
• There is no model improvement outside of custom learning.
• Complex integration into production applications

whisper

Released in September 2022, OpenAI’s Whisper is a cutting-edge open-source option. It supports multilingual transcription and can be used in Python or from the command line. Whisper offers five models of varying sizes and features.

merit

• Multilingual transcription
• Available in Python
• Available in 5 models

disadvantage

• An in-house research team is required for maintenance.
• It costs a lot of money to run
• Complex integration into production applications

Which free speech-to-text API, AI model, or open-source engine is right for your project?

The best free Speech-to-Text API, AI model, or open-source engine will depend on your project requirements. If ease of use, high accuracy, and additional features are your priorities, consider one of the APIs. However, if you prefer a completely free option with no data limits and are willing to put in extra work, an open-source library may be a better fit. Make sure the solution you choose can meet your current and future project requirements.

Image source: Shutterstock