Skip to main content
AssemblyAI Partner Service is available from CLI version 1.51.0 and greater
Cerebrium’s partnership with AssemblyAI helps teams deliver speech-to-text (STT) services with efficient deployment, minimized latency, and region selection for data privacy compliance needs.

​
Setup

  1. Contact the AssemblyAI team in order to get a self-hosted Contract. You can contact them here
  2. Create a simple cerebrium app with the CLI: 
cerebrium init assembly-ai
  1. AssemblyAI services use a simplified TOML configuration with the [cerebrium.runtime.assemblyai] section. Create a cerebrium.toml file with the following:
[cerebrium.deployment]
name = "assembly-ai"
disable_auth = true

[cerebrium.runtime.assemblyai]
port = 8080
model_name = "english"

[cerebrium.hardware]
cpu = 4
memory = 16
compute = "AMPERE_A10"
gpu_count = 1
region = "us-east-1"

[cerebrium.scaling]
min_replicas = 1
max_replicas = 3
cooldown = 120
replica_concurrency = 32
scaling_metric = "concurrency_utilization"
scaling_target = 70
The above disables auth meaning anyone can make requests to your endpoint. If you set disable_auth=false, then you need to use the API key from your Cerebrium Dashboard. key
  1. Run cerebrium deploy to deploy the AssemblyAI service - the output of which should appear as follows:
App Dashboard: https://dashboard.cerebrium.ai/projects/p-xxxxxxxx/apps/p-xxxxxxxx-assembly-ai
  1. Use the Deployment url from the output to send requests to the WS AssemblyAI service. We can use their example repo to test its working
1. git clone https://github.com/AssemblyAI/streaming-self-hosting-stack.git
2. cd streaming_example && python example_with_prerecorded_audio_file.py --audio-file example_audio_file.wav --endpoint wss://api.aws.us-east-1.cerebrium.ai/v4/p-xxxxxx/assembly-ai --language english
You should then see the following output: 
0:00:01.040000-0:00:01.200000, end-of-turn: False: it's true
0:00:01.040000-0:00:01.280000, end-of-turn: False: it's true that
0:00:01.040000-0:00:01.600000, end-of-turn: False: it's true that assem
0:00:01.040000-0:00:01.680000, end-of-turn: False: it's true that assembly
0:00:01.040000-0:00:02.080000, end-of-turn: False: it's true that assembly a
0:00:01.040000-0:00:02.160000, end-of-turn: False: it's true that assembly ai
0:00:01.040000-0:00:02.320000, end-of-turn: False: it's true that assembly ai lets
0:00:01.040000-0:00:02.400000, end-of-turn: False: it's true that assembly ai lets you
0:00:01.040000-0:00:02.560000, end-of-turn: False: it's true that assembly ai lets you build

​
Scaling and Concurrency

AssemblyAI services support independent scaling configurations:
  • min_replicas: Minimum instances to maintain (0 for scale-to-zero). Recommended: 1.
  • max_replicas: Maximum instances during high load.
  • replica_concurrency: Concurrent requests per instance. Recommended: 3.
  • cooldown: Seconds an instance remains active after last request. Recommended: 32.
  • compute: Instance type. Recommended: AMPERE_A10.
Adjust these parameters based on traffic patterns and latency requirements. Best would be to consult the Rime team about concurrency and scalability For further documentation on AssemblyAI, see the AssemblyAI documentation.