Spanish transcription turns Spanish speech into Spanish text. Before you compare services, write down the output you need: Spanish text, an English rendering, or both. A quote for one is not a quote for the others. Then test a short sample from your actual recording, especially if speakers change languages or the audio is noisy.
This guide compares one published human-service rate card, shows how Spanish locale and model availability differ, and explains why two often-cited accuracy numbers cannot be treated as a direct comparison. For related workflows, see our transcription service guide and multilingual audio guide.
- Landmark's published starting rates are $4.99 per audio minute for Spanish text, $6.99 for an English version, and $10.99 for both; extras may change the quote.
- Google Cloud's V2 model table differs by locale and region. A locale code is a support option, not an accuracy guarantee.
- A research review reports 48.38% word error rate on one Spanish-English code-switching benchmark. A separate model card reports 6.91% on Common Voice Spanish. They are not comparable product scores.
- MirrorCaption can transcribe live Spanish speech or an uploaded recording. Review important names, numbers, and quotations against the audio.
Compare the Deliverable Before the Rate
As checked in November 2026, Landmark lists three starting rates for the same Spanish recording. This is one agency's published price card, not a market average or a guaranteed final quote.
| Requested output | What you receive | Published starting rate |
|---|---|---|
| Spanish transcript | Spanish audio as Spanish text | $4.99 / audio minute |
| English version | Spanish audio rendered in English | $6.99 / audio minute |
| Both versions | Spanish text and English translation | $10.99 / audio minute |
Landmark says speaker tracking and timestamps are add-ons, while rush work can cost extra. Ask vendors to itemize the language of each deliverable, speaker labels, timestamps, review process, turnaround, and any certification. If you only need to search an internal interview, a machine draft with human spot-checking may be enough. If a filing needs an attested record, ask the provider about its human review and certificate.
Want a draft before requesting a human-service quote? Try your own Spanish audio and inspect the output.
Try MirrorCaption FreeCheck Locale and Model Availability
Spanish is spoken across many regions. In a speech API, choosing a locale can change which models are offered. Google Cloud Speech-to-Text V2 currently lists the following examples; availability also depends on deployment region and recognition method.
| Locale example | Listed model example | What to check |
|---|---|---|
| es-ES, es-US, es-MX | Chirp 3 in US/EU regions | Confirm the region and method you will use |
| es-419 | Chirp 2 in several regions | Test a Latin American sample |
| es-AR | Telephony and telephony_short in the listed regions | Try representative Argentine audio |
These entries describe Google's catalog, not MirrorCaption's underlying engine or a ranking of dialect quality. A phone-tuned model's presence does not prove that it will perform poorly on every room recording. Conversely, a newer model's name does not prove it will handle a particular accent, microphone, or code-switched conversation well. Compare models on the same short sample and count errors in terms, names, and numbers that matter to you.
Deepgram's model table illustrates a different way to label Spanish: its current Nova models list es and es-419, while some older models also list es-LATAM. Do not infer identical behavior from matching or differing locale labels across vendors.
Read Accuracy Claims in Their Test Context
A 2026 systematic review collected 127 peer-reviewed papers on end-to-end recognition of code-switched speech published through 2024. Its table records a 48.38% word error rate for a reported system on the spontaneous Spanish-English Miami Bangor corpus. The review also lists only three Spanish-English papers in its selected set. This is a result for a specific dataset and system, not the error rate of all Spanish audio or today's commercial services.
A Whisper fine-tuning model card reports 6.91% word error rate for its starting model on Common Voice 17 Spanish and 5.34% after fine-tuning. Common Voice uses prompted speech; the model card and the review use different systems, datasets, and evaluation setups. Dividing their percentages would produce a misleading accuracy comparison.
For a useful purchase test, take a representative clip from your own material. Include the hardest speaker, background noise, regional terms, and any Spanish-English switching. Compare each output with a checked transcript using the same scoring rules, then inspect high-impact errors by hand. Our accuracy comparison guide explains why the same audio can separate tools more clearly than advertised percentages.
Specify Your Spanish Transcription Job
- Define the output. Say whether you want Spanish text, an English translation, or both. Ask whether translation follows a reviewed Spanish transcript.
- Describe the audio. Share the recording length, speaker count, dialects, sound quality, and whether speakers alternate between Spanish and English.
- Choose the timing. Live captions help during a conversation; an uploaded recording produces a later transcript. See live versus post-recording transcription.
- State review requirements. Identify names, figures, quotations, timestamps, privacy needs, and any certificate your workflow requires. For EU recordings, review the data handling questions.
For example, a bilingual interview team may need Spanish quotations for its archive and an English summary for colleagues. That is different from needing two aligned, professionally translated transcripts. Writing the actual deliverable into the request avoids paying for the wrong output.
Where MirrorCaption Fits
MirrorCaption supports Spanish live captions and translation in the app. Its Transcriber accepts an existing recording and returns a transcript asynchronously. You can inspect and export text or Markdown, then check it against the audio. The app has no Spanish dialect selector, does not export SRT or VTT captions, and does not issue certified transcripts. A recording with frequent Spanish-English switching deserves an especially careful test; the upload flow uses one selected source language.
Try the free allowance on a representative clip before choosing a paid plan. Check current pricing and included credit in the app because plan terms can change. If you need a legally certified or fully human-reviewed record, request that service explicitly from a qualified provider.
Frequently Asked Questions
Is Spanish transcription the same as translation?
No. Transcription turns Spanish speech into Spanish text. Translation renders the meaning in another language. A quote should state whether you need the Spanish original, an English version, or both.
How much does Spanish transcription cost per audio minute?
Landmark lists starting rates of $4.99 per audio minute for Spanish text, $6.99 for an English version of Spanish audio, and $10.99 for both. Confirm add-ons, rush fees, and the current quote before ordering.
Does Spanish dialect affect transcription?
It can. Google Cloud Speech-to-Text V2 lists different models by Spanish locale and deployment region. That availability is not a measured accuracy score; test representative audio before choosing a model or vendor.
How accurate is automatic Spanish transcription?
There is no single accuracy number. A 2026 review reports 48.38% word error rate for one system on the code-switched Miami Bangor corpus; a separate model card reports 6.91% on Common Voice Spanish. These results use different systems and datasets and are not a head-to-head comparison.
Can MirrorCaption transcribe an existing Spanish recording?
Yes. Its Transcriber accepts an uploaded recording and returns the result asynchronously. Check the output against the audio before using it as a record; MirrorCaption does not provide certified transcripts or SRT/VTT export.
Test a Spanish Recording
Try a representative clip, review the transcript, and decide which files need human attention.
Get Started Free