Sold by: Murf Inc
Deployed on AWS
Speech-to-text for voice agents, contact centers and transcription in English, six European languages and 11 Indian languages. Finch streams transcripts while people speak, delivers each final transcript within about 250 ms of the end of speech, and runs inside your AWS account with network isolation, so audio never leaves your account.
Overview
Murf Finch converts speech to text from a live stream, a recorded file, or a batch of files in Amazon S3. Stream audio with InvokeEndpointWithBidirectionalStream and receive partial transcripts while the speaker talks and a final transcript at the end of each turn; or send a file with InvokeEndpoint and get the full transcript. Finch detects the spoken language automatically and can be biased toward key terms such as product, place and people names.
Who it is for: teams building voice agents and IVR, contact-center analytics and quality assurance, meeting and media transcription, and captioning, particularly where audio must stay inside their AWS account.
Versions: Finch Global transcribes English, German, Spanish, French, Italian, Portuguese and Dutch. Finch Indic transcribes Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Odia and Assamese, plus Indian English, with punctuation and word-level timestamps.
Accuracy: on the FLEURS test sets, Finch Global averages 4.9% word error rate across its seven languages (English 5.2%) and Finch Indic scores 10.3% on Hindi.
Latency and throughput, measured on Amazon SageMaker with live 16 kHz streams (November 2026): one ml.g5.2xlarge serves 96 concurrent streams and one ml.g6.xlarge serves 64, with 90% of final transcripts delivered within 250 ms of the end of speech. On ml.g6e.xlarge that figure is under 100 ms at 96 streams.
Model and training data: Finch is a neural speech recognition model built with PyTorch, trained on tens of thousands of hours of licensed and public multilingual speech.
Known limitations: Finch Global returns lowercase text without punctuation and does not yet return word timestamps. Speaker diarization is not included. Each version transcribes only its own languages. Accuracy is lower on noisy, overlapping or far-field speech; test with your own audio. Real-time requests are limited to 25 MB; use batch transform for files up to one hour.
Privacy: the model package runs with SageMaker network isolation, so no audio or text is sent to Murf or any third party. Real-time and streaming usage is billed per minute of audio; batch transform is billed per instance-hour.
Highlights
- Final transcripts within 250 ms of the end of speech (p90) at 96 concurrent live streams on one ml.g5.2xlarge, and under 100 ms on ml.g6e.xlarge.
- 4.9% average word error rate on FLEURS across English, German, Spanish, French, Italian, Portuguese and Dutch; 10.3% on Hindi, with 11 Indian languages plus Indian English in Finch Indic.
- Runs in your AWS account with network isolation. Live streaming, real-time files up to 25 MB and batch transform, with automatic language detection and key-term boosting, billed per audio minute.
Details
New
Introducing multi-product solutions
You can now purchase comprehensive solutions tailored to use cases and industries.
Features and programs
Financing for AWS Marketplace purchases
AWS Marketplace now accepts line of credit payments through the PNC Vendor Finance program. This program is available to select AWS customers in the US, excluding NV, NC, ND, TN, & VT.
Pricing
Pricing is based on actual usage, with charges varying according to how much you consume. Subscriptions have no end date and may be canceled any time.
Additional AWS infrastructure costs may apply. Use the AWS Pricing Calculator to estimate your infrastructure costs.
Dimension | Description | Cost |
|---|---|---|
ml.g6.2xlarge Inference (Batch) Recommended | Model inference on the ml.g6.2xlarge instance type, batch mode | $24.00/host/hour |
inference.count.m.i.c Inference Pricing | inference.count.m.i.c Inference Pricing | $0.003/request |
ml.g5.xlarge Inference (Batch) | Model inference on the ml.g5.xlarge instance type, batch mode | $24.00/host/hour |
ml.g6.xlarge Inference (Batch) | Model inference on the ml.g6.xlarge instance type, batch mode | $24.00/host/hour |
Vendor refund policy
Usage charges are non-refundable. For billing questions contact support@murf.ai .
Legal
Vendor terms and conditions
Upon subscribing to this product, you must acknowledge and agree to the terms and conditions outlined in the vendor's End User License Agreement (EULA) .
Content disclaimer
Vendors are responsible for their product descriptions and other product content. AWS does not warrant that vendors' product descriptions or other product content are accurate, complete, reliable, current, or error-free.
Delivery details
Amazon SageMaker model
An Amazon SageMaker model package is a pre-trained machine learning model ready to use without additional training. Use the model package to create a model on Amazon SageMaker for real-time inference or batch processing. Amazon SageMaker is a fully managed platform for building, training, and deploying machine learning models at scale.
Deploy the model on Amazon SageMaker AI using the following options:
Version release notes
Streaming session messages no longer include internal session settings. Adds ml.g6e.xlarge; ml.g5.2xlarge is now recommended (96 concurrent live streams). Updated guide: 25 MB real-time limit, getting started, Python streaming client.
Additional details
Inputs
- Summary
An audio file as the request body (Content-Type: audio/wav or application/octet-stream), with optional settings in CustomAttributes. Live streams take 16-bit PCM, mu-law, A-law or Opus audio frames.
- Limitations for input type
- 25 MB per real-time request. One audio file per batch transform input object, up to one hour long.
- Input MIME type
- audio/wav, application/octet-stream
https://raw.githubusercontent.com/murf-ai/murf-finch-sagemaker/main/samples/hi_in_sample.wav
https://raw.githubusercontent.com/murf-ai/murf-finch-sagemaker/main/samples/hi_in_sample.wav
Input data descriptions
The following table describes supported input data fields for real-time inference and batch transform.
Field name | Description | Constraints | Required |
|---|---|---|---|
audio | Audio file sent as the request body (WAV, FLAC, MP3, OGG/Opus, M4A and other common formats). | Up to 25 MB per real-time request; up to one hour per batch transform file. | Yes |
Custom attributes
The following table describes custom attributes for real-time inference endpoints.
Field name | Description | Constraints | Required |
|---|---|---|---|
language_code | Language of the audio, e.g. en, de, hi. Detected automatically when omitted. | One of the version's languages. | No |
timestamps_granularity | word returns word start and end times (Finch Indic); none omits them. | word or none | No |
keyterms | Words or phrases to recognize more reliably, separated by |. | - | No |
Support
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.
Similar products
Murf Falcon enables you to build voice agents that are ultra-fast, expressive, scalable, and highly cost-efficient. It has 55 ms model latency and 130 ms end-to-end latency. It supports 35+ languages and pricing starting at 1 cent per minute.
Murf Falcon 2.1 is a low-latency text-to-speech model for voice agents, IVR and narration. It streams the first audio in 44 ms (median, one stream on ml.g6e.xlarge) and speaks 42 locales with 141 voices, many of them in several languages. It runs entirely inside your AWS account with network isolation, so text and audio never leave your VPC.
Murf API offers enterprise-grade AI voice generation capabilities with its industry-leading text-to-speech models, transforming multimedia and conversational experiences. Our API achieves over 99% pronunciation accuracy and offers unparalleled speech customization through styles, pauses, duration matching and variation controls.
Deepgram is the enterprise Voice AI platform for building and scaling real time voice applications on AWS. This product listing contains multiple versions of the flux model which can each transcribe a set of languages. See version details for more information.
You will be billed $0.0077/min as described by https://deepgram.com/pricing. Private pricing available upon request.
Our APIs for Nova Speech to Text (STT) are natively available in the new SageMaker Bi-Directional Streaming API. Additional native touchpoints with Amazon Bedrock, Lex, and Amazon Connect make it simple to compose full voice experiences with the cloud services your teams already trust.
Customer reviews
No customer reviews yet
Be the first to review this product . We've partnered with PeerSpot to gather customer feedback. You can share your experience by writing or recording a review, or scheduling a call with a PeerSpot analyst.