Products

Located under Accent Configuration, account admins can choose to either run the model On-device or via Remote Inferencing.

Note: For more details, please visit Remote Inferencing.

Enable/disable translation independently for Outbound and Inbound audio.

Configure Accent Translation for the audio you send. To utilize this:

1. Set Source Accent and Target Accent

2. Select the Outgoing Accent Translation Model

3. Choose an Enhancement Mode

   1. Standard for a natural multi-speaker experience (keeps all voices within range), or

   2. Voice Isolation to filter out other voices

      1. Auto (automatic adjustment between Nearfield and General based on detected mic type)

      2. Nearfield (ideal for headset users when mic is directly at mouth position)

      3. General (works for all types of microphones)

Configure Accent Translation for the audio you receive. To utilize this:

1. Set Target Accent

2. Select the Incoming Accent Translation Model

3. Choose Incoming Audio Voice Activity Detection (VAD) Mode

   1. Voice Isolation - General with Ringtone Passthrough (works for all types of microphones while ensuring ringtones and hold music play normally), or

   2. Standard with Ringtone Passthrough to ensure ringtones and hold music play normally without being affected by Accent Translation processing

Note: By default, Target Accent is set and locked.

Choose what audio processing (for example, Noise Cancellation) applies when Accent Translation is toggled off.

If enabled, enforces a single voice profile (Masculine/Feminine) for users on first activation.

Toggle Enable Inbound Noise Cancellation to allow incoming-audio noise cancellation alongside outgoing Accent Translation.

Note: Select the Outgoing and Incoming Noise Cancellation model versions independently (each has a minimum required app version).

Located under Speech Enhancement Configuration, account admins can choose either run the model On-device or via Remote Inferencing.

Note: For more details, please visit Remote Inferencing.

Enable/disable independently for Outbound and Inbound audio.

Configure Speech Enhancement for the audio you send. To utilize this:

1. Choose an Enhancement Mode

   1. Standard for a natural multi-speaker experience (keeps all voices within range), or

   2. Voice Isolation to filter out other voices

      1. Auto (automatic adjustment between Nearfield and General based on detected mic type)

      2. Nearfield (ideal for headset users when mic is directly at mouth position)

      3. General (works for all types of microphones)

2. Select an Outgoing Speech Enhancement Model

Configure Speech Enhancement for the audio you receive. To utilize this:

1. Select an Incoming Audio Voice Activity (VAD) Mode

   1. Standard for a natural multi-speaker experience (keeps all voices within range), or

   2. Standard with Ringtone Passthrough to ensure ringtones and hold music play normally without being affected by Accent Translation processing

2. Choose an Incoming Speech Enhancement Model

Choose from the list of available languages in the dropdown menu as the default Source Language (what the App User speaks) and Target Language (what the recipient hears) in the app.

Note: App Users can change these default settings in the Sanas App.

When enabled, it shows users a real-time transcript of the conversation.

By default, this feature is enabled. For faster turn taking, you have the option to disable inbound speech for the agent.

Note: Once disabled, the agent will need to read the text from the translation and respond verbally.

Configure a custom glossary to prevent specific terms from being translated during multilingual calls. This is useful for industry-specific terminology, brand names, product names, or technical words that should be spoken as-is in the original language.

• Enable the Use Custom Glossary toggle to let users specify words or phrases that should be excluded from translation — such as client- or industry-specific terms, brand and product names, and business or technical jargon.

• Bulk upload terminologies via a .CSV file using the provided template. You can also download the current list, edit it, and re-upload to update entries in bulk.

Important Note: Add each term on its own, exactly as people say it on calls, such as product names, brand names, and acronyms. Avoid brackets or explanations in an entry. Entries like the ones below are matched as the exact phrase, brackets included, so they never come up in speech and have no effect. They can also cause the bracketed text to appear in the transcript.

• SSO (Single Sign On): add SSO and Single Sign On as separate entries instead.

• MFA (Multi-Factor Authentication): add MFA and Multi Factor Authentication as separate entries instead.

Bidirectional term substitutions are applied automatically during calls. Each entry maps one term to another and works in either direction — when set to Auto, the system uses whichever language the call originates in.

1. To start, turn the Use Custom Translations toggle ON.

2. For bulk terms upload, download the template, input your terms and translations, then upload.

3. For manual input of terms and translations, click +Add Translation.

4. Select two different languages from the dropdown menu, input the terms and translations, then click Save.

Note: Custom Translations is only available to Sanas App version 5.8.5 and above.

When enabled, Sanas admins will have the capability to debug translation issues faster and more efficient.

Delivers Language Translation transcripts, and optionally audio, to your own Amazon S3 bucket.

• How recordings are delivered?

  Per session. Each Language Translation session is delivered once, at the next daily run (03:00 UTC) after it ends.

• To start:

  • Let Sanas assume your IAM role

    • Create an IAM role in your AWS account and add the trust policy to it.

    • The external ID is unique to this group and stays the same.

  • Grant the role write access

  • Destination:

    • Files are delivered under <prefix>/<agentId>/<YYYY-MM-DD>/ in the bucket.

    • Fill out — S3 Bucket Name and AWS Region, then Path Prefix (Folder the files are delivered under), followed by IAM Role ARN (The role Sanas assumes to write into your bucket).

    • Test Connection (Assumes the role with this group's external ID and writes a small test file under the path prefix).

  • What to deliver

    • Transcript only (.txt)

    • Transcript and Audio (.txt, plus original and translated audio for the agent and the customer — four WAV files)

  Note: Unique IDs and scripts for Trust and Permission policies are available on the Sanas Portal.

When enabled, it displays measured translation latency on translated cards in the Sanas app. Effective for POC debugging and customer demos.

When enabled, it tunes the audio crossfading mixer in the Language Translation pipeline.

1. Choose a Crossfade Mode

   1. Server-side Crossfade (the translation server evens out the caller's voice and lowers it under the translation, most in the speech frequencies, before the audio reaches the app)

   2. Client-side Crossfade (the app mixes the caller's voice with the translation on the device)

      • Crossfade In/Out Timing

      • Activity Threshold

      • Low Gain

      • Ducking Level

   

Note: Advance Settings are for troubleshooting or demos only, typically for Sanas Admin use.

Option to select the appropriate models for the following component of Speech Intelligence:

• Speech to Text (STT)

  • Sanas’ on-device speech-to-text model that transcribes live calls locally on the agent’s machine and redacts PII before any data is sent, so the transcript can then feed summaries, action items and compliance alerts.

• Personally Identifiable Information (PII)

  • PII is personal information spoken on a call, such as names, account numbers and contact details, which Speech Intelligence automatically redacts on-device before any data is transmitted. Admins and compliance teams then review the PII-redacted transcriptions on the Speech Intelligence dashboard in the Sanas Portal.

• End of Call (EOC)

  • End of Call detection works with the Silence Detection model to recognize that a call has ended when it hears a farewell phrase like “bye” or “goodbye” followed by sustained silence from both participants. Once both conditions are met, Speech Intelligence automatically stops transcription and starts generating the post-call analysis and insights.

• Ringtone Classifier (RTC)

  • Ringtone Classifier is one of Speech Intelligence’s configurable models, alongside STT, PII and EOC, and admins can select its version per group from a picker in the model settings without a code change.

• Speaker Diarization (SEM)

  • Speaker Diarization is an optional model admins can pick on the Speech Intelligence card, either turned Off or set to a model with a speaker limit of 1 to 4, to separate who spoke when in the transcript. A group’s stored model only changes when an admin selects a new release, not when the card is opened.

Note: The STT and EOC models require Sanas App 5.5.5 or later, PII requires 5.9.0 or later, and the RTC model requires 5.10.0 or later.

Lets you choose the audio processing profile applied to the outgoing voice of your App Users, based on the type of microphone or call setup they're using.

Determines the audio processing profile applied to your customer or recipient’s voice on the call.

Enable the toggles for the sections you want to appear in generated call summaries in the Sanas app.

• Action Items

  • Show follow-up tasks identified during the call.

• Discussion Points

  • Show key topics and themes from the conversation.

• Compliance Alerts

  • Show potential risk or suspicious activity indicators.

• Comments (this is optional)

  • Show additional observations and manual inputs.

Note: Summary, Action Items, and Discussion Points are now configurable under Content Generation.

Controls how long and how much call transcript data is retained.

1. Percentage slider determines what portion of calls have their transcripts saved. At 0%, no transcripts are kept; below 100%, Sanas keeps a random sample of calls spread across the day; at 100%, every call is retained.

2. Retention Duration dropdown specifies how long transcripts are kept before deletion. After the selected period, transcripts are automatically hard-deleted by a background job (with up to ~6 hours of delay past the deadline), and this deletion is irreversible. Select an option from 1 day to 365 days.

3. Alert Utterance Retention toggle when enabled, lets you preserve transcript snippets tied to compliance alerts even after call transcript is deleted, so the alert details remain readable/reviewable later.

When enabled, delivers every completed call's redacted transcript in real time to an external destination.

• Once enabled, choose a Destination.

  1. Use existing destination

  2. Create destination

• Select Storage Type (Configured first, if needed)

  AWS (S3)

  Azure Blob

  To configure: Choose an External ID Pick any unique string (e.g. my-unique-externalId or a UUID). Use the same value in the trust policy and in Portal. This stops any other tenant from assuming your role. Create the IAM role Trust Policy (get the script from your Sanas POC) Permission Policy (get the script from your Sanas POC) Copy the role ARN after creation: arn:aws:iam::<your-account-id>:role/<your-role-name> Bucket Ownership Set Object Ownership → Bucket owner enforced so objects Sanas writes are owned by you. Enter Details in the portal S3 Bucket Name AWS Region Path Prefix Assume Role ARN External ID Click Test Connection Save only after it succeeds. Test Connection assumes your IAM role with the External ID you entered, and puts a small probe object under your prefix (overwrites; Sanas does not delete destination objects). Note: Sanas assumes an IAM role in your AWS account to transfer files to the configured S3 bucket. No long-lived access keys are stored in Portal.

  To configure: Create an Entra Application Microsoft Entra ID → App registrations → New registration. Name it (e.g. sanas-transcript-transfer). Single-tenant. Do not create a client secret or certificate. Copy the Application (client) ID and Directory (tenant) ID. Add a Federated Credential Certificates & secrets → Federated credentials → Add credential → Scenario Other Issuer Issuer / Subject / Audience = paste from Portal (exact match; no trailing slash); these are the sample values you should copy from the portal: Issuer: https://oidc.eks.us-east-1.amazonaws.com/id/<cluster-id> Subject: system:serviceaccount:sci-backend:sci-backend-prod-sa Audience: api://AzureADTokenExchange Name e.g. sanas-eks-prod Wrong subject or issuer is the usual “Connection failed” cause. Portal shows one generic write error for both token and permission failures. Storage Account + Container Create (or reuse) a storage account and blob container. Note both names for the Portal. RBAC — Storage Blob Data Contributor On the storage account or container → Access control (IAM) → Add Role Assignment Role: Storage Blob Data Contributor Assign to: User, group, or service principal → select the Entra App from Step 1 Review + Assign (allow a few minutes to propagate) Sanas only puts blobs under your prefix — it never lists, reads, or deletes destination objects. If the account uses a customer-managed key, also grant the app encrypt / unwrap on that key. Network If public access is disabled or a firewall is on, allow Sanas egress, or use private connectivity already arranged with Sanas. A blocked network looks like a write failure in Portal. Enter Details in the portal Tenant ID Client ID Storage Account Name Container Name Blob Prefix (optional) Click Test Connection Save only after it succeeds. Test Connection exchanges Sanas’ Kubernetes SA JWT for an Entra token for your app, and puts a small file to {prefix}.sci-connection-test/connection-test.txt (overwrites; never deleted by Sanas). Note: Sanas uses workload identity federation to write transcripts into your Azure Storage container. No client secret, SAS token, or account key is stored in the Portal. Also, there is no broker. Azure does not use an S3-style broker. You need (1) an Entra app with a federated credential that trusts Sanas, and (2) Storage Blob Data Contributor on the container for that app’s service principal.

  • AWS / Azure Mapping

    AWS (S3)	Azure
    IAM Role + External ID	Entra App + Federated Credential
    Sanas Principal ARN	Issuer + Subject + Audience from the Portal
    s3:PutObject	Storage Blob Data Contributor (put only in practice)
    Bucket + Prefix	Storage Account + Container + Optional Prefix
    KMS Key Policy	CMK Permissions for the App

  • Troubleshooting

    • For AWS S3:

      • Portal: “Connection failed…” — Check Trust Policy Principal + External ID match, s3:PutObject on bucket/*, and KMS key policy if the bucket uses SSE-KMS.

        • Trust — Principal is the Sanas IRSA role above; External ID matches Portal exactly

        • Permissions — s3:PutObject on the bucket objects ARN; KMS encrypt/generate, if needed

        • Names — bucket and region match Portal; prefix has no leading /

      • If still failing, send the following to Sanas Support:

        • Role ARN + External ID

        • Bucket + region + prefix

        • Trust + permission policy screenshots

        • UTC time of failed test

    • For Azure Blob:

      • Portal: “Connection failed. Sanas can’t write to this destination yet…” — generic. Check both federated credential and RBAC. “No broker” is not a valid skip.

        • Federated Credential — issuer/subject/audience match Portal; credential is on the same app as the Client ID in Portal

        • RBAC — Storage Blob Data Contributor on the app’s service principal, scoped to the storage account or named container.

        • Names / Encryption / Network — container exists and names match; CMK unwrap if used; firewall not blocking Sanas.

      • If still failing, send the following to Sanas Support:

        • Tenant ID + Client ID

        • Storage Account + Container

        • Federated Credential Screenshot

        • Role Assignment Screenshot

        • UTC Time of Failed Test

        • Entra Sign-in / Activity Log (if available)

• Metadata Transfer

  Once enabled, sends each call's metadata — call summary, action items, discussion points and compliance alerts — to the selected destination alongside the transcript. Sanas retains this metadata regardless of this toggle.

  Note: Delivered as a sidecar JSON object next to the transcript: {call_id}_{agent_id}_{YYYYMMDD}_metadata.json

Stream live transcripts to your own HTTPS endpoint, authenticated with an OAuth client-credentials credential. This is separate from the batch Transcript Transfer destinations.

1. Toggle the Redact PII ON.

2. Enter Delivery Endpoint URL (https address).

3. Fill out Authentication Credentials fields.

   1. Token URL

   2. Client ID

   3. Client Secret

   4. Optional (Scope and Audience)

4. Select the Destination.

   1. 1. Your Endpoint

      2. Sanas Cloud

      3. Both (Your Endpoint and Sanas Cloud)

         1. Both" sends each transcript to your endpoint AND stores a copy in Sanas cloud.

5. Click Test Connection.

   1. This checks that Sanas can reach your token URL and get a token with these credentials. It does not send a sample to your transcript endpoint.

Provides reusable, PII-redacted transcript datasets for testing Speech Intelligence configurations before applying them to live calls.

1. Dataset Library

   Upload your own dataset, sample real calls from production, or generate synthetic transcripts.

   

   Build Dataset through the following:

   1. Upload Dataset

      

   2. Sample from Production

      

   3. Generate Synthetic

      

2. Dataset Scripts

   Contains the list of datasets created for testing.

   

Narrative recap of what the call was about and how it ended.

To customize:

1. Click Configure.

2. From the pop-up window, choose among the following:

   1. Length (How long should each call summary be?)

      1. 1 to 2 sentences — under 30 words

      2. 3 to 5 sentences — 60 to 100 words

      3. Short paragraph — 100 to 150 words

   2. Tone (What tone should the summary use?)

      1. Formal / Business-Neutral

      2. Empathetic

   3. Point of View (Whose perspective should the summary take?)

      1. 3rd Person (Call POV)

      2. 3rd Person (Agent POV)

3. Select Review Changes to continue, then Prompt, and Test (optional) until completion.

Concrete future tasks that come out of the call.

To customize:

1. Click Configure.

2. From the pop-up window, choose among the following:

   1. Format (How should action items be laid out?)

      1. Bulleted List

      2. Structured Table

   2. Volume (How many action items per call?)

      1. Every Action Item

      2. Cap at 3

      3. Cap at 5

      4. Dynamic Cap

3. Select Review Changes to continue, then Prompt, and Test (optional) until completion.

What was talked about? The topics and notable moments of the call.

To customize:

1. Click Configure.

2. From the pop-up window, choose among the following:

   1. Format (How should discussion points be laid out?)

      1. Flat Bullets

      2. Grouped Sections

   2. Volume (How many discussion points per call?)

      1. Top 3

      2. Top 5

      3. All Meaningful Points

3. Select Review Changes to continue, then Prompt, and Test (optional) until completion.

Runs alerts on your production calls, with up to 10 active at once.

How to create a New Alert?

1. Click +New Alert tab.

2. Choose from the following, then Next:

   1. Start blank

   2. Start from template

3. Enter Alert Name.

4. Select Industry, Call Type, Detection Type, and Channel, then Next.

5. Click Create Alert.

   1. The Alert will be saved to Non-Production Library and gets an Alert Code.  The name, industry, call type or channel can’t be changed afterwards. Alerts can't be deleted, but can be archived.

6. Describe the behaviour to flag — the phrasing to require or prohibit, any exceptions, and timing windows.

   1. Ten chances to optimize the prompt.

7. Click Next to test the Alert before publishing, or at anytime within these steps, save it as a draft under Non-Production.

   

The Alert fields under Production defines the following:

Offers ready-made alerts for your industry and call type that you can fork and customize.

Define the Target Duration for Live Tracking supervisor timer in MM:SS format.

Note: The timer only accepts durations between 00:30 and 60:00. An amber alert triggers at 80% of the target duration, and a red alert triggers once the target is exceeded.

It controls which post-call sections, such as Call Stats, Summary, Action Items, and Compliance Alerts, appear in the Sanas Desktop App and the Agent Web Portal for a group. Turning a section off only hides it; everything is still generated and stored.

It controls whether compliance alerts run on calls the agent dials out. Turn on Suppress alerts on outbound calls to stop all in-call and post-call alerts on outbound calls for the group.

Note: Inbound calls, and calls whose direction is unknown, still get alerts as usual.

Select from the dropdown of Industry Type and Call Type.

Note: Downstream Speech Intelligence modules use this combination for defaults and filtering.

Enabling the Use Custom Glossary toggle activates the added terms, allowing Speech Intelligence models to recognize and transcribe them correctly.

To add the glossary terms:

1. Enter the words manually, separated by commas, or

2. Bulk upload them through .txt format.

Redaction Terms

• Mark customer terms for redaction with a tag such as MEDICAL or ORG.

  Note: One tag per line: TAG then the term. Redaction applies to new calls after the engine picks up the glossary. Calls already stored are not changed.